Characterizing the Growth in Utterance Length, Grammaticality, and Subordination in Spanish and English in Bilingual Children

Abstract

Purpose:

The aim of this study was to examine the growth of utterance length, grammaticality, and subordination in Spanish and English in bilingual children with and without developmental language disorder (DLD).

Method:

This is a longitudinal study of 100 children (DLD, n = 43) who were followed for 2 years with three testing times. At Time 1, children were, on average, at age 5;10 (years;months; SD = 12 months). Children produced spontaneous language samples in Spanish and English using a story retell and a story generation task. Language samples were transcribed and analyzed using the Systematic Analysis of Language Transcripts. Mean length of utterance in words (MLUw), percentage of grammatical utterances (PGU), and subordination index (SI) measures were calculated in Spanish and English. Growth curve modeling was used to analyze the data.

Results:

Similar growth trajectories were observed for MLUw and SI in both Spanish and English. These measures increased over time. However, PGU showed a different growth trajectory in Spanish and English. English PGU increased significantly, while Spanish PGU change was minimal. Children with DLD underperformed compared to children without DLD on all measures at the onset of the study, but their growth trajectories did not differ.

Conclusions:

Sentence development in bilingual children with and without DLD was characterized by increases in utterance length and use of subordination. However, grammaticality increased only in English. Clinicians should be aware of this developmental trend while conducting language evaluations in bilingual children.

The purpose of this study is to describe the growth in utterance length, grammaticality, and complexity in Spanish and English in bilingual children with and without developmental language disorder (DLD). DLD is a language-based learning disorder in which individuals present significant limitations in language ability that are not a result of sensory or neurological deficits (Bishop et al., 2016; Leonard, 2014). Children with DLD produce utterances that are shorter, less grammatical, and less complex than those produced by their peers with typical language development (Bedore & Leonard, 2005; Gutiérrez-Clellen et al., 2000; Restrepo, 1998). Importantly, bilingual children with DLD have production difficulties that manifest in all their languages (Paradis, 2005). Crucially, measures of grammaticality and utterance length are useful for the identification of bilingual children with DLD (Hernandez et al., 2024; Restrepo, 1998; Simon-Cereijido & Gutiérrez-Clellen, 2007). However, it is not yet well understood how these measures develop over time in bilingual children with and without DLD. A better understanding of how these measures develop over time would positively impact diagnostic practices for bilingual children in the United States.

In this study, we use three global indexes of morphosyntax extracted from spontaneous language samples. The first index is the mean length of utterance in words (MLUw). Mean length of utterance (MLU) is a well-studied measure of utterance length that has been traditionally calculated using morphemes in English (Brown, 1973; average number of morphemes per utterance). In Spanish, utterance length is better indexed using words to account for its rich morphosyntactic complexity (Gutiérrez-Clellen et al., 2000). Therefore, MLUw is used in both languages for comparability when examining sentence length in Spanish–English bilingual children. The second index is the percentage of grammatical utterances (PGU), which is a broad measure of grammaticality (Restrepo, 1998; number of grammatical utterances over all intelligible utterances produced by children multiplied by 100 to create a percentage). Our last index is the subordination index (SI), which is a measure of clausal density that provides broad information about syntactic complexity (Loban, 1963; total number of main and subordinated clauses over a total number of utterances). These global indexes are derived from spontaneous language samples, which are typically elicited through contexts such as free-play (child interacting with toys or materials with minimal adult direction), conversation (semistructured back-and-forth exchange between the child and an adult), or narratives (child retelling stories with the support of books; Miller & Iglesias, 2017).

In monolingual children, the development of sentence length (MLUw), grammaticality (PGU), and clausal density (SI) generally tends to occur in tandem (Brown., 1973). As children's utterance length increases (i.e., they produce more words within an utterance), the grammaticality of these utterances also tends to increase because children add more obligatory elements to the utterance (e.g., from “Frog jump” to “The frog jumped”). Subordination is observed with increases in utterance length (e.g., from “The frog is sad” to “The frog is sad because the boy left”) because children have more capacity to produce longer utterances or because the ability to produce complex sentences increases utterance length (Guo & Schneider, 2016). Although developmentally, a relationship between utterance length, grammaticality, and complexity seems to exist because they evolve in parallel, this relationship might not necessarily be mediated by development (i.e., these measures are correlated to age and, therefore, correlated to each other). Instead, these global measures might change independently of one another. Examining the development of MLUw, PGU, and SI in both languages of the bilingual child provides a unique opportunity to explore the growth of these measures of Spanish and English, which might be different across languages.

The languages of the bilingual child are developing due to both maturation (i.e., children are getting older) and language exposure (i.e., children vary in the exposure to Spanish and English over time). Importantly, in the U.S. context studied here, Spanish is often developing in conditions of reduced input and support, while English is rapidly growing because of increases in input and high support from the school and the community at large (e.g., Hiebert & Rojas, 2021; Lutz, 2008). Therefore, bilingual children in the United States tend to go through a shift in language dominance in which Spanish, traditionally their home or heritage language, becomes the nondominant language and English, the language of school and the larger community, becomes the dominant language (Castilla-Earls et al., 2019; Lutz, 2008). This shift in language dominance is not a result of a language learning limitation but instead is an effect of the sociolinguistic context that offers more support to one language than the other (Bills et al., 1995). The effects of these changes in input and support for Spanish and English might differ by measure (utterance length, grammaticality, and complexity) and/or by language, which has implications for the language development of children with and without DLD.

Development of MLU, PGU, and SI in Typically Developing Bilingual Children

Most longitudinal studies on global language measures with bilingual children have examined the growth patterns of MLUw and PGU in typically developing (TD) children. For MLUw, overall increases in Spanish and English have been described in the literature. Notably, Rojas and Iglesias (2013) examined the longitudinal development of MLUw using a large sample of narratives from TD bilingual children between 5 and 8 years of age. They found that MLUw increased in both Spanish and English, although growth in English seemed to slow down during the summer months, presumably when children spent more time at home. In a smaller study, Hiebert and Rojas (2021) examined the Spanish development of both MLUw and PGU using narratives in a sample of 35 TD children between the ages of 3 and 5 years. Their results suggested that Spanish MLUw increased over time, but Spanish PGU did not grow. Spanish PGU, in their study, plateaued at about 70%. However, they did not examine the developmental changes in English. Escobedo et al. (2023) examined the growth of PGU in Spanish and English in a small group of 4-year-old TD bilingual children using free-play. Children were tested at the start and the end of the school year. Their results suggested that PGU in English increased while PGU in Spanish did not change over time. Lucero (2018) examined longitudinal changes in MLUw and SI in both Spanish and English using narratives in a sample of 12 TD children from kindergarten to second grade. Their results suggested no changes in MLUw or SI for either language. Castilla-Earls et al. (2019) examined changes in PGU and MLUw over time in a large group of Spanish–English bilingual children using narratives. Their results suggested that in English, children increased both their MLUw and PGU over time. In contrast, in Spanish, increases in MLUw were accompanied by decreases in PGU.

In summary, most studies have used narratives for their examination of language growth. In English, the majority of studies find a pattern of increases in MLU and PGU over time. In Spanish, however, the growth patterns reviewed suggest increases in MLU but not in grammaticality, which tends to plateau or decrease. In contrast, the growth of SI, which indexes the development of subordination, in bilingual children has been understudied. Furthermore, the only study reviewed regarding the growth of SI suggests no changes over time in either Spanish or English. However, previous findings need to be expanded to examine the growth of MLUw, PGU, and SI together to examine sentence development in both Spanish and English to better describe the language development of bilingual children.

Language Growth in Children With DLD

Most of our knowledge regarding language development in children with DLD is based on monolingual English-speaking children. The current evidence suggests that these children have a delayed onset of language skills but similar growth trajectories compared to their TD peers (see Rice, 2013, for a review). Although children with DLD may appear to be catching up with their peers, this pattern seems to be measure-specific and reflects changes in the sensitivity of particular language tasks rather than language performance that is behaviorally indistinguishable from TD peers. For example, 5-year-old TD children outperform children with DLD on the correct production of finiteness, but by the age of 9 years, children with DLD catch up with their peers (Rice, 2013). Although this could be interpreted as a catching-up effect, this is more related to the fact that this skill levels out at 6 years of age in TD children. At 9 years of age, children with DLD are outperformed by TD children on other measures, such as grammatical judgments of finiteness (Rice, 2013), which suggests that the manifestations of DLD are persistent but the measures that capture this persistence change. There is limited information about the growth of language skills in bilingual children with DLD, but the current evidence suggests that bilingual children show the same pattern as monolingual children. That is, bilingual children with DLD show a delayed onset of the production of grammatical forms, and their growth pattern is similar to that of bilingual TD children in both languages (Castilla-Earls, Ronderos, & Fitton, 2023; Castilla-Earls, Ronderos, & Francis, 2023).

Regarding measures of sentence development, there is some information about the development of MLU in monolingual children with DLD. Children with DLD consistently perform below age expectations on MLU (Rice et al., 2006, 2010). Although monolingual English TD children produce utterances that are, on average, 3.5 words at 3 years of age, children with DLD do not reach this MLU until they are approximately 5 years of age (Rice, 2012). Importantly, MLU derived from free-play reached a natural leveling of five words in TD children, but children with DLD did not reach this performance level until around 8 years of age (Rice, 2012). To our knowledge, no studies have examined the growth of PGU or SI in children with DLD using a longitudinal design. However, Guo et al. (2020) used a cross-sectional approach to examine the effect of age on PGU in a group of TD children and children with language impairment. ¹ Their results suggested that there was a persistent gap in PGU performance between TD children and those with language impairment between the ages of 4 and 9 years. In both clinical groups, age group differences were observed, suggesting that PGU increased over time. Research on the development of MLUw, PGU, or SI in bilingual children with DLD is not available to the best of our knowledge.

This Study

Measures derived from spontaneous language samples are often used in clinical practice and research to examine the language development of children (Miller & Iglesias, 2017). In this study, we investigate the growth of MLUw, PGU, and SI using narratives in bilingual children with and without DLD. We ask the following research question: How do Spanish and English MLUw, PGU, and SI change over time in bilingual children with and without DLD? Based on previous research, we had the following predictions:

1. English MLUw, PGU, and SI will increase over time (Castilla-Earls et al., 2019; Escobedo et al., 2023; Rojas & Iglesias, 2013).

2. Spanish MLUw will increase, while PGU will plateau (Castilla-Earls et al., 2019; Escobedo et al., 2023; Hiebert & Rojas, 2021; Rojas & Iglesias, 2013).

3. Children with DLD will differ from TD children on their overall performance on MLUw, PGU, and SI in both languages but not on the rate of growth (Rice et al., 2006, 2010).

Method

Approval for this study was obtained from the institutional review board at the University of Houston (Study #973). This was a three time-point longitudinal study conducted between 2018 and 2021 to investigate language changes over time in bilingual children with and without DLD. Participants in this study are the same children described in Castilla-Earls, Ronderos, and Fitton (2023) and Castilla-Earls, Ronderos, and Francis (2023). Children in this study provided assent, and their parents provided consent for their children's participation. Time 1 data collection was conducted in person. Due to COVID-19, Time 2 data collection was conducted both in person and remotely. All children were subsequently tested remotely at Time 3 to avoid confounding testing modality and growth. For a discussion of the COVID-19–related adaptations in data collection, please refer to the method sections in Castilla-Earls, Ronderos, and Fitton (2023) and Castilla-Earls, Ronderos, and Francis (2023).

Participants

Children were recruited primarily from schools in the greater Houston metropolitan area but also through community outreach efforts. One hundred children qualified for this study. Children were, on average, at age 5;10 (years;months; SD = 1;0) at Time 1. Of the sample, 45 were female and 55 were male. Inclusion criteria included (a) exposure to both Spanish and English per parent report, (b) normal hearing, and (c) nonverbal intelligence that was within normal limits (> 70 on Kaufman Brief Intelligence Test–Second Edition Nonverbal subtest; Kaufman & Kaufman, 2004). Most children attended Spanish–English bilingual instructional programs (90%), Spanish–English bilingual schools (n = 75), dual language (n = 14), or Spanish immersion (n = 1); the remaining children attended English-only schools but had Saturday Spanish language school (n = 5).

All children lived in the Houston area, which has comparatively higher contextual support for Spanish than other U.S. regions. In terms of ethnicity and race, 97 children were reported to be Hispanic, one was reported to be non-Hispanic, and two did not respond; 71 were White/Caucasian, 16 were “Other,” and one was Mexican American. Parents were from Mexico (n = 52), United States (n = 27), Puerto Rico (n = 1), or another country not listed (n = 18). There was one no response (n = 1). In terms of the mother's highest level of education, there was a range from elementary (n = 27), high school (n = 30), some college (n = 9), associate degree (n = 7), bachelor's degree (n = 10), to graduate degree (n = 14).

To determine group assignment (i.e., DLD or TD) at Time 1, we used a converging evidence approach (Castilla-Earls et al., 2020) using the same criteria as in previous studies conducted with this sample (for details, see Castilla-Earls, Ronderos, & Fitton, 2023; Castilla-Earls, Ronderos, & Francis, 2023). Specifically, children were considered DLD if they met two of three clinical indicators: (a) They are currently receiving speech-language services. (b) Their performance is at or below the age-based cutoff on standardized language assessments indicating DLD (Bilingual English Spanish Assessment [BESA] Morphosyntax subtest [Peña et al., 2018], Bilingual English Spanish Assessment–Middle Extension [BESA-ME] Morphosyntax subtest [Peña et al., 2008], and Clinical Evaluation of Language Fundamentals Sentence Repetition subtest [Semel et al., 2006; Wiig et al., 2013]) when their BESA/BESA-ME confidence interval included the age-based cutoff score. The diagnostic accuracy of the BESA Morphosyntax was considered appropriate for the identification of children with DLD in this study (sensitivity is 89.7–96.5 and specificity is 83.2–89.9, depending on the age group). (c) Their performance is at or below the cutoffs, indicating DLD on measures of grammaticality and utterance length in both Spanish and English from a story retell task (Hernandez et al., 2024; Restrepo, 1998). The use of these criteria resulted in 43 children in the DLD group and 57 children in the TD group. The overrepresentation of DLD in this sample was intentional because we targeted children receiving speech-language services; this is not a normative sample of the population. Table 1 provides descriptive information for the participants in this study.

Table 1.

Descriptive statistics for participants' demographic information and language scores.

Variable, M (SD)	Time 1 (N = 100)			Time 2 (N = 90)			Time 3 (N = 81)
	TD	DLD	p	TD	DLD	p	TD	DLD	p
Participants	57 (57.0%)	43 (43.0%)		53 (58.9%)	37 (41.1%)		49 (60.5%)	32 (39.5%)
Age in months	72.02 (11.71)	67.49 (11.95)	.061	87.51 (13.35)	81.51 (13.43)	.039	96.41 (12.18)	90.16 (12.58)	.029
Sex, n (%)
Male	26 (45.6%)	29 (67.4%)	.030	25 (47.2%)	25 (67.6%)	.055	21 (42.9%)	21 (65.6%)	.045
Female	31 (54.4%)	14 (32.6%)		28 (52.8%)	12 (32.4%)		28 (57.1%)	11 (34.4%)
Home language(s)
Spa only	32 (56.1%)	23 (53.5%)	.792
Spa and Eng	25 (43.9%)	20 (46.5%)
Receiving services
Yes	5 (8.8%)	41 (95.3%)	< .001
No	52 (91.2%)	2 (4.7%)
Nonverbal IQ	103.23 (11.80)	101.91 (15.48)	.629	103.43 (11.95)	102.92 (15.40)	.859	103.06 (12.02)	103.47 (15.68)	.895
BESA/BESA-ME Spa	91.42 (18.66)	71.84 (10.26)	< .001	85.72 (20.75)	68.03 (14.29)	< .001	90.87 (21.77)	69.38 (17.06)	< .001
BESA/BESA-ME Eng	92.32 (16.89)	68.83 (10.17)	< .001	100.28 (14.43)	73.39 (14.03)	< .001	103.31 (13.65)	80.03 (13.83)	< .001
TVIP	94.18 (17.31)	79.77 (14.87)	< .001	93.85 (21.21)	78.54 (17.23)	< .001	98.00 (21.18)	79.78 (17.13)	< .001
PPVT	89.05 (19.34)	73.79 (14.07)	< .001	92.89 (20.54)	76.50 (16.63)	< .001	98.91 (19.95)	83.06 (13.33)	< .001

Note. For all language measures, standard scores are reported. TD = typically developing; DLD = developmental language disorder; Spa = Spanish; Eng = English; Nonverbal IQ = Kaufman Brief Intelligence Test–Second Edition (Kaufman & Kaufman, 2004); BESA = Bilingual English–Spanish Assessment (Peña et al., 2018); BESA-ME = Bilingual English-Spanish Assessment–Middle Extension (Peña et al., 2008); TVIP = Test de Vocabulario en Imágenes; PPVT = Peabody Picture Vocabulary Test.

Procedure

All children were administered a battery of language assessments in Spanish and English on two separate days by two different bilingual research assistants. In each session, examiners exclusively used the target language and encouraged the child to respond in that language. The tasks of interest in this study were the story retell and story generation tasks. For these tasks, we used four stories: two in English, Frog Goes to Dinner (Mayer, 1974) and Frog on His Own (Mayer, 1973), and two in Spanish, A Boy, a Dog, and a Frog (Mayer, 1967) and One Frog Too Many (Mayer & Mayer, 1975). Within each language, children were offered the two books assigned to the target language and asked to choose one of the books. Children were then asked to listen to the story while looking at the pictures and then to retell the story to the research assistant. We used the story scripts provided by the Systematic Analysis of Language Transcripts (SALT) website at the time the study was conducted. Once the retell was complete, the child looked at the second book and was asked to tell a story while looking at the pictures. Research assistants were trained to use strategies to help the child retell and generate stories, such as silently turning the pages as children were producing language, asking questions to elicit language (e.g., “What else happened?”), or instructing the child to produce language (e.g., “Tell me what happened here”).

Children's stories were audio-recorded and subsequently transcribed using SALT (Miller & Iglesias, 2017) by trained bilingual research assistants. Language samples followed SALT transcription conventions. The transcription process was conducted using a systematic approach to increase rigor. First, one Spanish–English bilingual research assistant completed an initial transcription of the audio recording using SALT. Second, another bilingual research assistant reviewed the transcription while listening to the audio and marked any disagreement. Finally, a third research assistant resolved disagreements between the first and second transcribers. Therefore, the final transcription consisted of a language sample in which there was agreement between two transcribers for each word (Heilmann et al., 2008).

Measures

MLUw

Language samples were segmented using modified C-units to ensure comparability across Spanish and English. Unlike traditional C-unit procedures, modified C-units segment coordinated utterances with the same subject as belonging to different utterances (e.g., The frog jumped/And looked for the boy). Notably, complex–compound utterances (e.g., Cuando sacó lengua de la flor y se comió la abeja, la abeja lo picó [When he stuck his tongue out from the flower and ate the bee, the bee stung him]) were considered a single C-unit to capture syntactic complexity. Following SALT conventions, Spanish clitic pronouns were treated as separate words in both enclitic and proclitic positions. MLUw was automatically output in both languages using SALT.

PGU

Utterances that contained more than a single noun, adjective, or interjection (e.g., Frog. Sad. Yeah.) were coded for grammaticality. Each utterance was evaluated independently to determine its grammaticality, consistent with previous studies (Heilmann et al., 2008). We distinguished grammatically from felicity (i.e., semantically and/or pragmatically appropriateness) and considered as grammatical any well-formed utterance that could plausibly relate to the shared context (i.e., frog story pictures and storyline). This approach ensured that children were not penalized for ambiguous referents or marginally infelicitous contexts (e.g., a sentence with a singular referent when multiple referents were expected based on the story). In other words, children were not penalized for lexical or pragmatic choices as long as the overall morphosyntactic structure was grammatically well formed.

We followed a rigorous protocol to complete grammaticality coding. The grammaticality of each utterance and the site of errors in production were assessed through a coding process conducted in each language, although we only report on the overall utterance-level grammaticality in this study. After identifying production errors, the grammaticality of each intelligible utterance was determined by marking it as [G] (i.e., grammatical) or [UG] (i.e., ungrammatical). The occurrence of these codes per participant was extracted from SALT to calculate PGU for English and Spanish separately by totaling all grammatical utterances (i.e., [G]), dividing by the total number of intelligible utterances in the analysis set (i.e., [G] + [UG]), and multiplying by a hundred. Coding reliability was established through a consensus method, where all utterances coded as grammatical and ungrammatical were reviewed by all authors; disagreements were discussed until a consensus was achieved by at least three authors. This approach was used to add rigor to the grammaticality coding since agreement was required by at least three coders.

The following grammaticality coding principles were followed:

1. Dialectal considerations. Spanish-speaking children in the United States are exposed to various regional and social varieties. Thus, care was taken to avoid penalizing nonnormative and/or stigmatized forms in lexical, morphosyntactic, and phonological domains. Lexical variation included common nouns (e.g., “grass” as zacate, pasto, grama, or césped), as well as regional forms that are stigmatized (e.g., nadien/nadie [no one], namás/nada más [only], onde/dónde [where]), and common truncated forms (e.g., garró/agarró [got], taba/estaba [was]). Morphosyntactic variation included second-person singular preterit –s (e.g., com-istes [you ate]). Phonological variation is discussed in point (d). All these differences were accepted as correct.

2. Bilingual grammars. As other grammars, bilingual grammars are evolving, and procedures should consider the sociolinguistic context of these speakers to avoid penalizing children for typical patterns seen in heritage Spanish contexts. Therefore, instances of zero-marked differential object marking (Ve el niño vs. Ve al niño [S/he sees the boy]) or periphrastic future (e.g., Voy ir vs. voy a ir [I'm gonna go]) were not penalized.

3. Code switching. Utterances containing a single-noun insertion or insertion of one simple conjunction (e.g., and la rana brincó [the frog jumped]) in the other language were included in the analyses and coded for MLUw, PGU, and SI. We accepted as grammatical any Spanish determiner that was used with the code-switched English noun (e.g., el frog, la frog). All other code switches (e.g., phrasal, intrasentential, dense) were excluded from the analyses (see Table 2 for details on excluded utterances due to other code switches).

4. Phonological differences. Words with phonological differences were considered grammatical if intelligibility was not impacted. These included developmental errors (e.g., cluster reduction: produced: bincó/ target: brincó [s/he jumped]), bilingual patterns showing cross-linguistic influence (e.g., rhotic production or voice onset time differences in voiceless stops /p, t, k/), or phonological variation such as /s/ aspiration (ejcuela/escuela [school]) and neutralized liquids (puelta/puerta [door]).

5. Verb-specific considerations. Verb root alternations in stem-changing verbs were accepted as correct (e.g., dormió/durmió, juegar/jugar), while paradigmatic errors were considered incorrect and given word error codes; these included verbal suffixes with overregularized verb forms (e.g., comiba/comía [eat. imperfect], ponió/puso [put.preterite], caye/cae [falls]) or conjugation class errors (e.g., vestando/vistiendo [getting dressed (conjugated as first conjugation instead of third conjugation)]).

6. Semantic errors. Infelicitous utterances (i.e., those that were semantically unexpected) were tagged with a word- or utterance-level semantic code [s] and further subcharacterized into grammatical and ungrammatical instances, as exemplified below.

Table 2.

Descriptive statistics for language sample measures.

Variable, M (SD)	Time 1 (N = 100)			Time 2 (N = 90)			Time 3 (N = 81)
	TD	DLD	p	TD	DLD	p	TD	DLD	p
Spanish
Utt Num	60.93 (25.69)	47.98 (26.28)	.016	66.45 (27.71)	61.50 (27.50)	.413	73.67 (28.03)	71.22 (25.16)	.693
Unintell Utt Excl	3.95 (4.72)	14.19 (18.37)	< .001	3.49 (4.74)	11.72 (11.83)	< .001	2.59 (3.71)	8.25 (8.84)	< .001
CS Utt Excl	3.50 (10.67)	7.14 (13.74)	.143	5.18 (17.26)	5.56 (12.33)	.910	8.25 (8.84)	4.28 (7.37)	.048
MLUw	6.29 (1.32)	4.67 (1.20)	< .001	7.09 (1.13)	5.93 (1.23)	< .001	7.55 (1.08)	6.50 (1.22)	< .001
PGU	68.51 (21.36)	46.21 (18.37)	< .001	73.65 (17.26)	61.16 (17.23)	.002	71.32 (20.03)	54.85 (20.53)	< .001
SI	1.21 (0.14)	1.05 (0.15)	< .001	1.27 (0.15)	1.18 (0.10)	.003	1.32 (0.15)	1.17 (0.14)	< .001
English
Utt Num	58.96 (25.14)	32.98 (27.69)	< .001	62.23 (19.59)	44.68 (26.69)	< .001	68.39 (19.68)	62.77 (33.43)	.366
Unintell Utt Excl	2.53 (5.43)	10.67 (11.45)	< .001	4.29 (4.88)	15.35 (15.37)	< .001	2.80 (4.89)	9.53 (12.10)	.001
CS Utt Excl	2.59 (10.13)	16.58 (24.29)	< .001	0.58 (3.21)	8.19 (17.65)	.003	0.11 (0.62)	2.87 (7.45)	.017
MLUw	6.70 (1.35)	4.66 (1.73)	< .001	7.20 (1.47)	5.92 (1.29)	< .001	7.76 (0.80)	6.70 (1.19)	< .001
PGU	58.25 (23.21)	31.97 (20.48)	< .001	69.87 (22.91)	47.19 (20.99)	< .001	77.04 (21.22)	51.45 (19.75)	< .001
SI	1.12 (0.15)	0.83 (0.31)	< .001	1.20 (0.19)	1.01 (0.21)	< .001	1.26 (0.12)	1.13 (0.16)	< .001

Note. TD = typically developing; DLD = developmental language disorder; Utt Num = number of utterances; Unintell Utt Excl = number of utterances excluded based on unintelligibility; CS Utt Excl = number of utterances excluded based on certain types of code switching; MLUw = mean length of utterance in words; PGU = percentage of grammatical utterances; SI = subordination index.

The grammatical semantic codes included lexical substitutions, verb pairs, and calques:

• Lexical substitutions were noun/adjective substitutions that were semantically related (e.g., produced: rat when target was frog; mosca when target was abeja), phonologically related (e.g., ovispa/avispa), or neologisms (e.g., musiquero/músico, salada/ensalada).

• Three semantically close verb pairs in both languages with the same argument structure (come and go; look for and find; take and bring).

• A small set of English calques in Spanish was accepted as grammatical: (estar como or fue como for decir; preguntar for pedir; estar frío for tener frío; estaba for the existential verb haber; jugar + instrument for tocar).

The ungrammatical semantic codes were copula selection errors and argument structure errors:

• Copula selection was substitution of one copula verb for another (e.g., ser for haber or for estar); the sole exception was the calque mentioned above.

• Verbs with incorrect argument structure (e.g., eat milk) were coded as ungrammatical.

SI

SI coding was done using SALT's “Insert SI code” function. The third author, first author, and a research assistant, all bilingual and with expertise in Spanish and English linguistics, reviewed the SALT “SI Scoring Guide” and developed training materials that included additional rules to ensure reliability. Following SALT coding for SI, abandoned or unintelligible utterances and those utterances with the [EU] utterance code (i.e., unrecoverable meaning) were excluded from SI count. Utterances with more than 50% in the nontarget language, segment fragments, elliptical structures in response to questions, and meta-comments/off-topic remarks (e.g., “I don't know this one”) were also excluded from analysis using the [SI-X] code.

All other utterances were coded according to the number of clauses in the utterance, defined as the number of main verbs. The [SI-1] code included simple sentences with a single tensed verb (e.g., eats) or compound verb forms with auxiliaries or modals (e.g., had finished, was scolded, will eat, could go, is gonna go). Utterances that were full propositions but did not have a clear verb (e.g., sound effects or verb/copula omission) were considered ungrammatical zero clauses, coded as [SI-0]. For English only, this also included subject omission that was not the result of segmentation rules. Given the polysemy of so as a coordinating or subordinating conjunction, all instances of so were extracted with their preceding and following utterance to determine if its use was consistent with a coordinating function, in which case it was coded as two separate utterances or instead a purposive subordinating conjunction.

Significant departures from the SALT manual are treatment of infinitival structures and specification of ungrammatical structures. Infinitival to– (He wants to play; She wants him to come) and bare infinitival clauses (He saw him play) were considered biclausal [SI-2], following previous work (e.g., Diessel, 2004; Gutierrez-Clellen, 1998). We developed a protocol to ensure that children were not penalized for nontarget morphosyntactic errors while coding for subordination. A guiding principle for determining whether clauses under consideration showed evidence of subordination was that clauses with added or substituted obligatory functional elements were counted, while those with omitted obligatory elements were not (e.g., Y dice *que no tiene miedo [And s/he says *that s/he is not scared]).

SI reliability was conducted for English and Spanish on a random 20% of transcripts (English: n = 54, Spanish: n = 55). Reliability was calculated as number of agreements divided by the total number of agreements plus disagreements. In English, utterance-by-utterance reliability yielded 97.7% agreement. In Spanish, agreement was 95.1%. These reliability estimates for English and Spanish were considered appropriate. All disagreements were discussed and resolved.

The definition of the analysis set for all transcripts was the same for English and Spanish. Specifically, the analysis set included all intelligible utterances obtained from the combination of the story retell and generation tasks that were coded with either [G] or [UG] per participant per language per time point to ensure the three measures were calculated from the same analysis set each time (see Table 2 for details on excluded utterances due to unintelligibility). Using the Rectangular Data File tool in the Research Version of SALT, we extracted the standardized measures (i.e., including MLUw and SI) and the number of instances of the codes [G] and [UG]. PGU was calculated as previously described from the extracted data per participant per language per time point.

Data Analysis Plan

We used growth curve modeling using mixed models to examine growth trajectories for each outcome variable in Spanish and English measures of utterance length (MLUw), grammaticality (PGU), and subordination (SI). In order to examine Spanish and English at the same time in a single measurement model, we employed a hyperunivariate format for data, where outcome scores for each outcome type (MLUw, PGU, and SI) were stacked under a single observation uniquely identified by participant ID, language (Spanish or English), and participant's age in months (using a similar approach as used in Castilla-Earls et al., 2019; Castillla-Earls, Ronderos & Fitton, 2023). Models were estimated using the MIXED command in Stata/BE Version 18 (Stata, 2023). For each observation of an individual child, we replicated time-invariant covariates, such as sex and mother's education, while time-variant covariates, such as age, were unique for each record of a given participant. A series of growth curve models for each outcome variable (MLUw, PGU, and SI) were estimated to determine the best-fitting mixed model that evaluated the fixed effects of the variables we hypothesized would be associated with growth over time for these measures, namely, age and language ability classification status and the interaction between them. Random effects for participant, language, and time were included in the models to allow for intercept and slope variability of selected variables at each step. Models were constructed using the following incremental steps.

The first model (Model 1) estimated the unconditional means model, which did not include the effect of time. This unconditional model was the initial step for our incremental analyses to describe and partition outcome variation in English and Spanish within the same model. Furthermore, this model included estimates for Spanish and English intercepts with random effects for participant and language and allowed for separate variance residuals for each language. This allows the model to account for variation at the participant level, while also allowing Spanish and English to vary independently from each other within each participant. The equation in Model 1 was defined by Equation 1.

$${\mathit{\text{Outcome}}}_{\mathit{il}}={\gamma }_{00l}+{\cup }_{0\mathit{il}}+{\in }_{\mathit{il}}$$ (1)

where Outcome_il is the predicted level of each outcome (MLUw, PGU, or SI) for language l (Spanish and English) and participant i. Therefore, in this model, γ_00l is the language-specific intercept and ∪_0il is the participant-specific intercept per language, which varies randomly between participants, and ∈_il is the random error residual at the participant and language level.

Model 2 aimed to further explain the variability in the outcome variables by adding age (in months) as a predictor to model growth over time and to allow the slope for age to randomly vary across participants. This model was used to answer whether the outcome measures (MLUw, PGU, and SI) vary as a factor of the participant's age and how old the participant is at the first time point of the study. Additionally, Model 2 allows Spanish and English growth to be modeled separately within a participant. The equation for Model 2 is defined in Equation 2.

$${\mathit{\text{Outcome}}}_{\mathit{itl}}={\gamma }_{00l}+{\cup }_{0\mathit{il}}+{\gamma }_{10l}\times {\stackrel{—}{\mathit{age}}}_{{\mathit{\text{child}}}_{\mathit{it}}}+{\gamma }_{20l}\times {\stackrel{—}{\mathit{age}}}_{{\mathit{\text{group}}}_{\mathit{it}}}+{\gamma }_{30l}\times {\stackrel{—}{\mathit{age}}}_{{\mathit{\text{child}}}_{\mathit{it}}}\times {\stackrel{—}{\mathit{age}}}_{{\mathit{\text{group}}}_{\mathit{it}}}+{\cup }_{1\mathit{il}}\times {\stackrel{—}{\mathit{age}}}_{{\mathit{\text{child}}}_{\mathit{it}}}+{\in }_{\mathit{itl}}$$ (2)

Age was included in Model 2 using the ${\stackrel{—}{\mathit{age}}}_{\mathit{\text{child}}}$ and ${\stackrel{—}{\mathit{age}}}_{\mathit{\text{group}}}$ variables that were calculated separately. The first variable, ${\stackrel{—}{\mathit{age}}}_{\mathit{\text{child}}}$, was the age for each participant i centered at the onset of the study (i.e., age for the child at each time point minus age for the child at Time 1). Including this variable in the model allows for the examination of age over time for each participant (i.e., the age for each participant in comparison to themselves at different ages). The second variable, ${\stackrel{—}{\mathit{age}}}_{\mathit{\text{group}}}$, was the average age of all participants in the study centered at the onset of the study (i.e., the age of the child at each time point minus the average age of all children at Time 1). This second variable allows for the prediction of age over time between the participants in the study (i.e., the age for each participant in comparison with the average age for participants at Time 1 of the study). Model 2 also includes the interaction between ${\stackrel{—}{\mathit{age}}}_{\mathit{\text{child}}}$ and ${\stackrel{—}{\mathit{age}}}_{\mathit{\text{group}}}$, which is used to account for differences in growth rates that are due to participant's age at the onset of the study. Therefore, the intercept in Model 2, γ_00l, is interpreted as the language-specific intercept for the outcome (MLUw, PGU, or SI) at the onset of the study for a participant of average age (based on the group) at the onset of the study. In Model 2, ∪_0il continues to be the participant-specific intercept per language (reflecting between-participant variation), ∪_1il is the participant-specific slope (over time) per language, and ∈_il is the random error residual at the participant and language level.

Model 3 adds to the variability explained by Model 2 by introducing language ability classification status (i.e., the participant is categorized as DLD by having DLD = 1 or as TD by having DLD = 0) in the model. By adding language ability classification in Model 3, we can answer the research question of whether children with DLD have Spanish and English outcomes for these measures (MLUw, PGU, and SI) that are significantly different from their TD peers. The equation for Model 3 is defined in Equation 3.

$${\mathit{\text{Outcome}}}_{\mathit{itl}}={\gamma }_{00l}+{\cup }_{0\mathit{il}}+{\gamma }_{10l}\times {\stackrel{—}{\mathit{age}}}_{{\mathit{\text{child}}}_{\mathit{it}}}+{\gamma }_{20l}\times {\stackrel{—}{\mathit{age}}}_{{\mathit{\text{group}}}_{\mathit{it}}}+{\gamma }_{30l}\times {\stackrel{—}{\mathit{age}}}_{{\mathit{\text{child}}}_{\mathit{it}}}\times {\stackrel{—}{\mathit{age}}}_{{\mathit{\text{group}}}_{\mathit{it}}}+{\gamma }_{40l}\times {\mathit{DLD}}_{{\mathit{\text{child}}}_{\mathit{it}}}+{\cup }_{1\mathit{il}}\times {\stackrel{—}{\mathit{age}}}_{{\mathit{\text{child}}}_{\mathit{it}}}+{\in }_{\mathit{itl}}$$ (3)

Model 3 added a coefficient for the prediction of a child having DLD. Thus, the intercept for Model 3, γ_00l, now represented the language-specific intercept for the outcome (MLUw, PGU, or SI) for a TD participant language, (i.e., DLD = 0) who was of average age at the onset of the study.

The final model, Model 4, expanded on Model 3 by adding the interaction between language ability classification status and the child's age (${\stackrel{—}{\mathit{age}}}_{\mathit{\text{child}}}$). Adding the interaction allows to answer the research question whether children with DLD have different growth rates compared to their TD peers. The equation for Model 4 is defined in Equation 4.

$${\mathit{\text{Outcome}}}_{\mathit{itl}}={\gamma }_{00l}+{\cup }_{0\mathit{il}}+{\gamma }_{10l}\times {\stackrel{—}{\mathit{age}}}_{{\mathit{\text{child}}}_{\mathit{it}}}+{\gamma }_{20l}\times {\stackrel{—}{\mathit{age}}}_{{\mathit{\text{group}}}_{\mathit{it}}}+{\gamma }_{30l}\times {\stackrel{—}{\mathit{age}}}_{{\mathit{\text{child}}}_{\mathit{it}}}\times {\stackrel{—}{\mathit{age}}}_{{\mathit{\text{group}}}_{\mathit{it}}}+{\gamma }_{40l}\times {\mathit{DLD}}_{{\mathit{\text{child}}}_{\mathit{it}}}+{\gamma }_{50l}\times {\mathit{DLD}}_{{\mathit{\text{child}}}_{\mathit{it}}}\times {\stackrel{—}{\mathit{age}}}_{{\mathit{\text{child}}}_{\mathit{it}}}+{\cup }_{1\mathit{il}}\times {\stackrel{—}{\mathit{age}}}_{{\mathit{\text{child}}}_{\mathit{it}}}+{\in }_{\mathit{itl}}$$ (4)

This model would predict the age over time within participants with and without DLD. Model fit between models was evaluated using the Akaike information criterion (AIC) and the likelihood ratio test (LRT). AIC evaluates the relative quality of how a statistical model fits the data by taking into account the complexity of the model when assessing model fit. Smaller values for the AIC indicate better model fit (Bandalos & Finney, 2019). The LRT allows for comparison of model fit between models at a chosen confidence level (i.e., p < .05).

Results

Descriptive statistics for language sample measures by age are summarized in Table 2. Fixed and random effects of all models for all outcomes of interest (MLUw, PGU, or SI) are presented in Tables 3–8. We describe the results of the best-fitting model for all outcome measures.

Table 3.

Fixed effects for utterance length (mean length of utterance in words [MLUw]) growth in Spanish and English.

Model	Spanish			English			Difference between Spanish and English estimates
	Estimate	SE	p	Estimate	SE	p
Model 1
MLUw	6.30	0.13	< .001	6.48	0.14	< .001	0.338
Model 2
MLUw	5.61	0.12	< .001	5.74	0.14	< .001	< 0.001
${\stackrel{—}{\mathit{age}}}_{\mathit{\text{child}}}$	0.06	0.01	< .001	0.06	0.01	< .001	0.998
${\stackrel{—}{\mathit{age}}}_{\mathit{\text{group}}}$	0.06	0.01	< .001	0.07	0.01	< .001	0.374
${\stackrel{—}{\mathit{age}}}_{\mathit{\text{child}}}\times {\stackrel{—}{\mathit{age}}}_{\mathit{\text{group}}}$	< 0.001	< 0.001	.036	< 0.001	< 0.001	.012	< 0.001
Model 3
MLUw	6.08	0.14	< .001	6.34	0.15	< .001	< 0.001
${\stackrel{—}{\mathit{age}}}_{\mathit{\text{child}}}$	0.06	0.01	< .001	0.06	0.01	< .001	0.997
${\stackrel{—}{\mathit{age}}}_{\mathit{\text{group}}}$	0.05	0.01	< .001	0.06	0.01	< .001	0.405
${\stackrel{—}{\mathit{age}}}_{\mathit{\text{child}}}\times {\stackrel{—}{\mathit{age}}}_{\mathit{\text{group}}}$	< 0.001	< 0.001	.018	< 0.001	< 0.001	.002	0.190
DLD	−1.06	0.19	< .001	−1.41	0.21	< .001	0.222
Model 4
MLUw	6.20	0.15	< .001	6.49	0.16	< .001	< 0.001
${\stackrel{—}{\mathit{age}}}_{\mathit{\text{child}}}$	0.05	0.01	< .001	0.05	0.01	< .001	0.936
${\stackrel{—}{\mathit{age}}}_{\mathit{\text{group}}}$	0.05	0.01	< .001	0.06	0.01	< .001	0.392
${\stackrel{—}{\mathit{age}}}_{\mathit{\text{child}}}\mathit{\times }{\stackrel{—}{\mathit{age}}}_{\mathit{\text{group}}}$	< 0.001	< 0.001	.047	< 0.001	< 0.001	.010	0.183
DLD	−1.36	0.23	< .001	−1.79	0.25	< .001	0.202
DLD $\mathit{\times }{\stackrel{—}{\mathit{age}}}_{\mathit{\text{child}}}$	0.03	0.01	.183	0.03	0.01	.025	0.733

Note. Bold font indicates the best-fitting model. SE = standard error; DLD = developmental language disorder.

Table 4.

Random effects for utterance length (mean length of utterance in words [MLUw]) growth in Spanish and English.

Model	Spanish		English		AIC
	Estimate	SE	Estimate	SE
Model 1
Level 1	1.24	0.14	1.54	0.18	1,833.62
Level 2	1.05	0.24	1.19	0.29
Model 2
Level 1	0.69	0.08	0.98	0.12	1,631.68
Level 2 int	0.86	0.17	1.01	0.22
Level 2 slope	< 0.001	< 0.001	< 0.001	< 0.001
Model 3
Level 1	0.69	0.09	0.96	0.11	1,570.82
Level 2 int	0.55	0.14	0.57	0.15
Level 2 slope	< 0.001	< 0.001	< 0.001	< 0.001
Model 4
Level 1	0.66	0.09	0.93	0.11	1,562.33
Level 2 int	0.60	0.14	0.57	0.14
Level 2 slope	< 0.001	< 0.001	< 0.001	< 0.001

Note. Bold font indicates the best-fitting model. SE = standard error; AIC = Akaike information criterion.

Table 5.

Fixed effects for grammaticality (percentage of grammatical utterances [PGU]) growth in Spanish and English.

Model	Spanish			English			Difference between Spanish and English estimates
	Estimate	SE	p	Estimate	SE	p
Model 1
PGU	63.73	1.91	< .001	55.67	2.30	< .001	0.007
Model 2
PGU	60.33	2.06	< .001	46.34	2.34	< .001	< 0.001
${\stackrel{—}{\mathit{age}}}_{\mathit{\text{child}}}$	0.30	0.07	.023	0.77	0.10	< .001	< 0.001
${\stackrel{—}{\mathit{age}}}_{\mathit{\text{group}}}$	0.31	0.17	.024	0.51	0.20	.068	0.460
${\stackrel{—}{\mathit{age}}}_{\mathit{\text{child}}}\times {\stackrel{—}{\mathit{age}}}_{\mathit{\text{group}}}$	< 0.001	0.01	.712	< 0.001	0.01	.505	0.649
Model 3
PGU	67.54	2.43	< .001	55.99	2.61	< .001	< 0.001
${\stackrel{—}{\mathit{age}}}_{\mathit{\text{child}}}$	0.30	0.07	.015	0.78	0.10	< .001	< 0.001
${\stackrel{—}{\mathit{age}}}_{\mathit{\text{group}}}$	0.18	0.16	.194	0.33	0.18	.126	0.530
${\stackrel{—}{\mathit{age}}}_{\mathit{\text{child}}}\mathit{\times }{\stackrel{—}{\mathit{age}}}_{\mathit{\text{group}}}$	0.00	0.01	.681	0.00	0.01	.407	0.501
DLD	−16.60	3.50	< .001	−22.77	3.82	< .001	0.234
Model 4
PGU	69.10	2.53	< .001	56.60	2.69	< .001	< 0.001
${\stackrel{—}{\mathit{age}}}_{\mathit{\text{child}}}$	0.16	0.10	.334	0.71	0.12	< .001	< 0.001
${\stackrel{—}{\mathit{age}}}_{\mathit{\text{group}}}$	0.14	0.16	.274	0.32	0.18	.148	0.459
${\stackrel{—}{\mathit{age}}}_{\mathit{\text{child}}}\times {\stackrel{—}{\mathit{age}}}_{\mathit{\text{group}}}$	< 0.001	0.01	.927	0.01	0.01	.348	0.544
DLD	−20.30	3.90	< .001	−24.41	4.23	< .001	0.475
DLD $\times {\stackrel{—}{\mathit{age}}}_{\mathit{\text{child}}}$	0.34	0.15	.194	0.18	0.20	.546	0.542

Note. Bold font indicates best-fitting model. SE = standard error; DLD = developmental language disorder.

Table 6.

Random effects for grammaticality (percentage of grammatical utterances [PGU]) growth in Spanish and English.

Model	Spanish		English		AIC
	Estimate	SE	Estimate	SE
Model 1
Level 1	153.63	17.19	285.97	33.22	4,480.23
Level 2	295.58	51.17	401.35	76.78
Model 2
Level 1	139.30	15.59	157.03	32.52	4,401.60
Level 2 int	289.93	49.51	379.29	72.13
Level 2 slope	< 0.001	< 0.001	0.22	0.17
Model 3
Level 1	137.04	18.51	167.43	31.95	4,357.35
Level 2 int	226.61	40.99	246.91	52.21
Level 2 slope	0.01	0.07	0.17	0.15
Model 4
Level 1	134.35	17.88	165.80	31.69	4,355.83
Level 2 int	228.87	40.97	248.06	52.34
Level 2 slope	< 0.001	0.06	0.17	0.15

Note. Bold font indicates best-fitting model. SE = standard error; AIC = Akaike information criterion.

Table 7.

Fixed effects for subordination index (SI) growth in Spanish and English.

Model	Spanish			English			Difference between Spanish and English estimates
	Estimate	SE	p	Estimate	SE	p
Model 1
SI	1.198	0.013	< .001	1.088	0.020	< .001	< 0.001
Model 2
SI	1.144	0.014	< .001	0.981	0.021	< .001	< 0.001
${\stackrel{—}{\mathit{age}}}_{\mathit{\text{child}}}$	0.005	0.001	< .001	0.009	0.001	< .001	< 0.001
${\stackrel{—}{\mathit{age}}}_{\mathit{\text{group}}}$	0.004	0.001	< .001	0.010	0.002	< .001	0.004
${\stackrel{—}{\mathit{age}}}_{\mathit{\text{child}}}\times {\stackrel{—}{\mathit{age}}}_{\mathit{\text{group}}}$	< 0.001	< 0.001	.980	< 0.001	< 0.001	.047	0.004
Model 3
SI	1.195	0.016	< .001	1.064	0.023	< .001	< 0.001
${\stackrel{—}{\mathit{age}}}_{\mathit{\text{child}}}$	0.005	0.001	< .001	0.009	0.001	< .001	< 0.001
${\stackrel{—}{\mathit{age}}}_{\mathit{\text{group}}}$	0.003	0.001	.002	0.009	0.002	< .001	0.004
${\stackrel{—}{\mathit{age}}}_{\mathit{\text{child}}}\times {\stackrel{—}{\mathit{age}}}_{\mathit{\text{group}}}$	< 0.001	< 0.001	.951	< 0.001	< 0.001	.014	< 0.001
DLD	−0.116	0.022	< .001	−0.196	0.032	< .001	0.040
Model 4
SI	1.205	0.018	< .001	1.092	0.024	< .001	< 0.001
${\stackrel{—}{\mathit{age}}}_{\mathit{\text{child}}}$	0.004	0.001	< .001	0.006	0.001	< .001	0.080
${\stackrel{—}{\mathit{age}}}_{\mathit{\text{group}}}$	0.003	0.001	.003	0.009	0.002	< .001	0.004
${\stackrel{—}{\mathit{age}}}_{\mathit{\text{child}}}\mathit{\times }{\stackrel{—}{\mathit{age}}}_{\mathit{\text{group}}}$	< 0.001	< 0.001	.851	< 0.001	< 0.001	.058	0.008
DLD	−0.140	0.027	< .001	−0.269	0.038	< .001	0.006
DLD $\mathit{\times }{\stackrel{—}{\mathit{age}}}_{\mathit{\text{child}}}$	0.002	0.001	.309	0.006	0.002	.003	0.058

Note. Bold font indicates the best-fitting model. DLD = developmental language disorder.

Table 8.

Random effects for subordination index (SI) growth in Spanish and English.

Model	Spanish		English		AIC
	Estimate	SE	Estimate	SE
Model 1
Level 1	0.015	0.002	0.029	0.003	−281.00
Level 2	0.012	0.003	0.029	0.006
Model 2
Level 1	0.012	0.001	0.017	0.002	−419.89
Level 2 int	0.009	0.002	0.026	0.005
Level 2 slope	< 0.001	< 0.001	< 0.001	< 0.001
Model 3
Level 1	0.012	0.001	0.017	0.002	−472.40
Level 2 int	0.006	0.002	0.016	0.004
Level 2 slope	< 0.001	< 0.001	< 0.001	< 0.001
Model 4
Level 1	0.012	0.001	0.016	0.002	−483.47
Level 2 int	0.006	0.002	0.017	0.004
Level 2 slope	< 0.001	< 0.001	< 0.001	< 0.001

Note. Bold font indicates the best-fitting model. AIC = Akaike information criterion.

Utterance Length (MLUw)

Model 4, the model that included language ability and the interaction of language ability and child age, was the model that best fit the data to predict MLUw. Model 4 had the lowest AIC of all models for MLUw (1,562.33), and the LRT test showed that it was a significant improvement in fit compared to Model 3, χ²(2) = 12.49, p = .002 (all models for MLUw are summarized on Tables 3 and 4). Model 4 indicated that the average MLUw for TD children of average age at the onset of the study (i.e., at age 5;10) was 6.20 in Spanish (p < .001) and 6.49 in English (p < .001), which was a statistically significant difference (p < .001). The slopes for age for Spanish and English MLUw were positive, indicating an increase of 0.05 in MLUw per month in both languages (i.e., growth of 0.60 words per year; p < .001), which was not statistically different between the languages (p = .936). This suggests that the growth of MLUw in TD children does not differ for the two languages. Between-children differences in age (i.e., ${\stackrel{—}{\mathit{age}}}_{\mathit{\text{group}}}$ ) were significant in predicting MLUw in both languages, so that when children were 1 month older when starting the study, MLUw was greater by 0.05 for Spanish (p < .001) and by 0.06 in English (p < .001). The between-children difference was not significant between the languages (p = .392). The interaction between ${\stackrel{—}{\mathit{age}}}_{\mathit{\text{child}}}$ and ${\stackrel{—}{\mathit{age}}}_{\mathit{\text{group}}}$ was statistically significant at −0.001 for Spanish (p < .001) and −0.002 for English (p = .010). This interaction indicates a small decrease in the within-child growth rate for utterance length in English and Spanish when children are older at the start of the study.

Model 4 also included language ability and the interaction between language ability and age, which tests for differences in intercept and rate of growth between children with DLD and their TD peers. Children classified as having DLD at the onset of the study obtained significantly lower MLUw than their TD peers by 1.36 in Spanish (p < .001) and by 1.79 in English (p < .001). Thus, the average MLUw for children with DLD who were at age 5;10 at the onset of the study was 4.84 for Spanish and 4.70 for English, which were not statistically different between the languages (p = .202). The interaction between language ability and age was significant for English by 0.03 (p = .025) but not for Spanish (p = .183). This suggests that for children with DLD, English MLUw increases faster by an additional 0.03 per month (i.e., an additional 0.36 in MLUw per year) compared to their TD peers, while Spanish MLUw grows at the same rate as for their TD peers. This difference in growth of MLUw between the languages, however, was not statistically significant (p = .733).

The fixed effects for Model 4 predicting utterance length in Spanish and English for children with and without DLD are shown in Figure 1. The individual growth linear trajectories for MLUw predicted by Model 4 (i.e., fixed effects and random effects) are displayed in Figure 2. English is displayed on the left panel, and Spanish is displayed on the right panel. The figure summarizes the key takeaways for the growth of MLUw over time: Children tend to start with lower utterance length in Spanish than in English, and children with DLD produce significantly shorter utterances compared to their TD peers. The growth rate for Spanish and English tends to be about the same for both languages, except for children with DLD, whose utterance length tends to grow more rapidly in English than TD children.

Figure 1.

Predicted growth in utterance length over time by language (fixed effects based on Model 4). MLUw = mean length of utterance in words; Spa = Spanish; Eng = English; TD = typically developing; DLD = developmental language disorder.

Figure 2.

Individual predicted growth in utterance length over time by language (fixed and random effects based on Model 4). MLUw = mean length of utterance in words; TD = typically developing; DLD = developmental language disorder.

Grammaticality (PGU)

Model 3, the growth model that included fixed effects for language ability, was considered the best model for predicting grammaticality (i.e., PGU) in both languages. Model 4, which also included the interaction of language ability and child age, had an AIC of 4,355.83, compared to Model 3 (AIC = 4,357.35). However, the LRT test comparing Model 3 and Model 4 was not statistically significant, χ²(2) = 5.51, p = .064, suggesting that adding the interaction in Model 4 did not provide better data fit since the interaction was not significant. Fixed effects for all models predicting grammaticality are summarized in Table 5, and random effects for all models are detailed in Table 6.

Results from Model 3, the best-fitting model for grammaticality, indicate that TD children who were of average age at the onset of the study had, on average, 67.54% grammaticality in Spanish (p < .001) and 55.99% grammaticality in English (p < .001). The difference in intercepts between Spanish and English was statistically significant (p < .001), indicating that TD children were significantly more grammatical at age 5;10 in Spanish compared to English. The slope for grammaticality in both languages was positive, indicating an increase of 0.30% in PGU per month in Spanish (i.e., growth of 3.6% in PGU per year; p = .015) and 0.78% per month or 9.36% per year (p < .001) in English. The difference in grammaticality growth in Spanish and English was statistically significant (p < .001), indicating that developmental trajectories of grammaticality significantly differ for the two languages, with steeper growth for English. Between-children differences in age were not significant (i.e., Spanish: ${\stackrel{—}{\mathit{age}}}_{\mathit{\text{group}}}$ = 0.18, p < .001; English: ${\stackrel{—}{\mathit{age}}}_{\mathit{\text{group}}}$ = 0.33, p < .001), suggesting that being a month older at the beginning of the study did not significantly impact grammaticality in either language. Similarly, the interaction between ${\stackrel{—}{\mathit{age}}}_{\mathit{\text{child}}}$ and ${\stackrel{—}{\mathit{age}}}_{\mathit{\text{group}}}$ was not statistically significant for grammaticality (i.e., Spanish: p = .681, English: p = .407), and these effects were not significantly different for Spanish compared to English (i.e., p = .530 for ${\stackrel{—}{\mathit{age}}}_{\mathit{\text{group}}}$ differences between Spanish and English; p = .501 for the difference between the languages in the interaction between ${\stackrel{—}{\mathit{age}}}_{\mathit{\text{child}}}$ and ${\stackrel{—}{\mathit{age}}}_{\mathit{\text{group}}}$).

Since Model 3 included language ability in the model, it provided separate intercepts for children with DLD. Children with DLD were significantly less grammatical than their TD peers by 16.60% in Spanish (p < .001) and by 22.77% in English (p < .001). Therefore, the average grammaticality for children with DLD who were 5;10 at the onset of the study was 50.94% for Spanish and 33.21% for English. The difference between the effect of DLD in Spanish and English grammaticality was not statistically significant (p = .234). Figure 3 displays the fixed effects for Model 3 predicting grammaticality in Spanish and English for children with and without DLD. Individual growth curves for PGU predicted by Model 3 (i.e., fixed effects and random effects) are shown in Figure 4. One of the key takeaways for PGU growth over time is that children tend to start with lower grammaticality in English than in Spanish. However, the growth rate for grammaticality in English is much faster than in Spanish. Consistently across both languages, children with DLD are less grammatical than their TD peers, but their growth is at the same rate as that of TD children.

Figure 3.

Predicted growth in grammaticality over time by language (fixed effects based on Model 3). PGU = percentage of grammatical utterances; Spa = Spanish; Eng = English; TD = typically developing; DLD = developmental language disorder.

Figure 4.

Individual predicted growth in grammaticality over time by language (fixed and random effects based on Model 3). PGU = percentage of grammatical utterances; TD = typically developing; DLD = developmental language disorder.

SI

Model 4, the growth model that included fixed effects for language ability and the interaction between language ability and age, was considered the best model for predicting SI, similar to the utterance length results. The AIC for Model 4 was −483.47 compared to −472.40 for Model 3. Results from the LRT test comparing Models 3 and 4 were statistically significant, χ²(2) = 15.07, p < .001, suggesting that adding the separate growth rates for children with and without DLD, with the interaction term in Model 4, provided significantly better data fit. Fixed effects for all models predicting the SI in both languages are summarized in Table 7, and random effects for all models are detailed in Table 8.

TD children of average age at the onset of the study, per results from Model 4, had an average SI of 1.21 in Spanish (p < .001) and 1.09 in English (p < .001), which was a statistically significant difference between the languages, p < .001. This indicates that TD children had significantly more complexity in Spanish compared to English at the start of the study. Growth for the SI in Spanish was 0.004 per month (or 0.045 per year; p < .001), while for English, the SI increased by 0.006 per month (or 0.103 per year; p < .001), but the difference in growth rates was not statistically significant between the two languages (p = .080). An additional month of age at the onset of the study resulted in an additional 0.003 in the SI in Spanish (p = .003) and 0.009 in English (p < .001), and this was significantly different between the languages (p = .004). The estimated effect of the interaction between ${\stackrel{—}{\mathit{age}}}_{\mathit{\text{child}}}$ and ${\stackrel{—}{\mathit{age}}}_{\mathit{\text{group}}}$ was not statistically significant for SI in Spanish (i.e., < 0.001, p = .851) and English (i.e., < 0.001, p = .058), which indicates no significant change in the within-child growth rate for subordination in Spanish and English in older children at the onset of the study compared to younger children.

Children classified as having DLD had significant lower SI compared to their TD peers by 0.140 in Spanish (p < .001) and by 0.269 in English (p < .001). The average SI for children with DLD who were 5;10 at the onset of the study was 1.07 for Spanish and 0.82 for English. This difference between the SI in Spanish and English for children with DLD was statistically significant (p = .006). Furthermore, the interaction effect between language ability and age on the SI was significant for English by 0.006 (p = .003) but not significant for Spanish (p = .309). This result suggests that the SI in children with DLD increases by an additional 0.006 per month (i.e., an additional 0.077 in SI per year) compared to TD children but that growth in Spanish is similar for children with and without DLD. This difference in slope between English and Spanish SI production in children with DLD was not statistically significant (p = .058). The fixed effects for Model 4 predicting the SI are shown in Figure 5. Individual growth curves for SI predicted by Model 4 (i.e., fixed effects and random effects) are displayed in Figure 6. Results for the SI analyses suggest that children tend to start with a greater SI in Spanish than in English, and children with DLD produce less complex utterances compared to their TD peers, especially in English. The growth rate for Spanish and English tends to be about the same for both languages. However, as with utterance length, the SI tends to grow more rapidly in English for children with DLD than for TD children.

Figure 5.

Predicted growth in subordination index over time by language (fixed effects based on Model 4). SI = subordination index; Spa = Spanish; Eng = English; TD = typically developing; DLD = developmental language disorder.

Figure 6.

Individual predicted growth in subordination index over time by language (fixed and random effects based on Model 4). SI = subordination index; TD = typically developing; DLD = developmental language disorder.

Discussion

The purpose of this study was to examine the growth of MLUw, PGU, and SI in Spanish and English in children with and without DLD. There are two main findings in this study. First, the results of this study suggested different growth patterns among measures and between languages. In English, all measures examined in this study showed a pattern of increase over time. In Spanish, growth was evident in MLUw and SI. PGU also showed an increase in Spanish over time, but this change was very small. Second, children with DLD were outperformed by TD children at the onset of the study in all measures, but their growth rate was similar to the growth of TD children. This study is unique in its description of the developmental trajectories of Spanish and English in bilingual children with and without DLD, but its results must be interpreted with consideration for the study context.

This study was conducted in Houston, Texas, a city with notable community support for Spanish. For example, Spanish is spoken by 40% of the population in the city, and bilingual education programs are available and mandated for children who need to further develop their English skills. Furthermore, there is a strong presence of Latine/o culture and community, with representation from various countries, such as Mexico (73%), Honduras (4%), Guatemala (2%), El Salvador (8%), and Colombia (2%), among others (11%; U.S. Census Bureau, 2019). Therefore, Spanish is often spoken at various places in the community, such as stores, pharmacies, banks, and libraries. Although it is still the case that English holds higher relative prestige (e.g., Mazzaro & González de Anda, 2024), the strong presence of Spanish in the community provides unique opportunities to support the development of Spanish. Therefore, our results are not generalizable to bilingual children living in contexts with varying levels of support for their Spanish development. In the next sections, we discuss our results for Spanish and English growth and differences between children with and without DLD.

MLUw, PGU, and SI Growth Patterns in English

Our results present a unique description of English sentence development that includes the examination of utterance length, grammaticality, and complexity over time. Our models predict that a TD bilingual child of average age at the onset of the study (5;10) would produce stories with sentences that are about 6.5 words in length on average, with an SI of 1.10 and with about 55% of their utterances being grammatical. The low English grammaticality at the onset is a result of the language learning process for these bilingual children and not an indication of a disorder. That is, these children are in the process of acquiring English, and low grammaticality is part of the learning process. Children may produce multiword utterances that might include some early appearing subordinating conjunctions, such as “because” or “when,” and still produce utterances that are not 100% grammatical. It is important to highlight that these children's sentences in English increase over time for all measures (length, grammaticality, and complexity) examined in this study.

The growth observed in this study for English spontaneous language measures replicates the results of previous studies for MLUw (Castilla-Earls et al., 2019; Rojas & Iglesias, 2013) and PGU (Castilla-Earls et al., 2019; Escobedo et al., 2023), showing that both of these measures increase over time in bilingual children. Unlike Lucero (2018), we observed growth for SI and MLU in English; however, this may be a sample size effect, as their study included a small sample of 12 children. Children's growth in English, across all measures, was not surprising since English is the language of higher prestige in the community, and most of these children attend bilingual education programs that support both their English and Spanish development. The impact of the overall English-supporting environment on children's development of English is evident for children in this context.

MLUw, SI, and PGU Growth Patterns in Spanish

Our models predicted that a TD child of average age at the onset of the study (5;10) would produce stories with sentences that are approximately 6.2 words in length on average, with an SI of 1.20, and with about 68% of their utterances being grammatical. Measures of sentence length and subordination significantly increased over time, showing evidence of growth. Although our measure of grammaticality increased over time, the rate of growth was very small (a predicted change of approximately +5% between 48 and 90 months) and more in line with findings of language maintenance. In other words, these children maintained their overall grammaticality over time.

The continued growth in Spanish MLUw found in this study is in agreement with the results of Rojas and Iglesias (2013), Hiebert and Rojas (2021), and Castilla-Earls et al. (2019). Our results differ from Lucero (2018), who did not find a change in MLUw or SI over time. The small rate of growth effect observed for PGU in Spanish is somewhat in agreement with the results of Escobedo et al. (2023) and Hiebert and Rojas (2021), who found plateauing effects. Overall, our results align with previous results for Spanish MLUw and PGU and offer new insights into the development of SI.

The children in this study were in an environment that continued to support their Spanish, as evidenced by growth in all measures. Why would SI and MLU grow significantly while PGU shows minimal growth? These children continued to produce sentences with more words and greater subordination. Nevertheless, their grammaticality in Spanish improved very little. This is likely not an effect of input, with children receiving less Spanish input over time, since we would expect that all measures to show a similar pattern, which was not the case. Instead, we consider it more likely that there is an interaction between input and similarities/differences between the languages. As children grow, they receive less input in Spanish and more input in English, even in contexts where bilingual education is available (Lutz, 2008). However, the children in this study were receiving input in both languages and were attending schools that provided Spanish support. We hypothesize that MLUw and SI in both languages support each other because of the similarities between the languages. That is, the development of the two languages might provide support for each other, such that adding words to the utterance in one language might also add the ability to add words to the utterance length in the other. This might also be the case for subordination: Once the child uses subordination strategies in one language, the child will likely be able to implement subordination in the other. For example, if a child is using clauses of reason with the subordinated conjunction “because” in Spanish, it is likely they would attempt to use clauses of reason in English, provided they have enough English knowledge to create utterances. Crucially, syntactic complexity in Spanish and English is similar (e.g., conditionals and nominals) with some exceptions (e.g., questions and negation), so they might develop closely between the languages. Questions and negation are not included in the SI we used in this study. Furthermore, it is not necessary for all sentences to be similar between the languages to be able to support each other during development. Also, it is important to note that the way we code for SI does not penalize for grammatical errors because a child might produce a sentence that clearly shows subordination but that has elements that are not grammatical (e.g., the frog jumped right before the boy catched him). Grammaticality, on the other hand, may not share many properties between the languages, which may require more input for growth. Because the morphological systems differed between Spanish and English (e.g., verb morphology, grammatical gender), more input in Spanish might be needed for continued grammatical development in Spanish. It is possible that input in Spanish and English together support sentence development in both languages for utterance length and complexity but not for grammaticality because there are features that are shared between the languages (sentence structure and subordination) and features that are less shared (language-specific nominal and verbal morphology). Perhaps, with higher support for Spanish, we would see PGU in both languages develop in tandem. For example, PGU in Spanish grew at a very low rate in this study, with the majority of children attending bilingual education programs. In comparison, PGU in Spanish plateaued for the children in English-only schooling in Hiebert and Rojas (2021). In conditions of reduced input, language development increases in length and subordination, while grammaticality is maintained. However, more research is needed to better understand these patterns of language growth in bilingual children.

It is important to note that we did not observe a pattern of Spanish language loss for these children when we examined these global measures. MLUw and SI continued to grow, and PGU was maintained. This is likely an effect of the support for Spanish in the context of the Houston area. Although it may seem contradictory to our previous hypothesis, we must also consider that support for language development can vary among communities. Although Houston provides more Spanish support than other areas in the United States, some areas, such as border towns, offer even more contextual support for Spanish development. Therefore, the moderate support provided in the Houston area might result in Spanish maintenance for grammaticality, while areas with more support will see strong growth in all areas, including PGU. Alternatively, areas with minimum support for Spanish would probably result in Spanish loss.

Furthermore, our results do not suggest a change in language dominance for these children until about age 7.5, since Spanish measures, on average, are very similar or better than English at all points in time. The shift in language dominance from Spanish to English is a well-documented phenomenon among bilingual children in the United States (e.g., Castilla-Earls et al., 2019; Lutz, 2008). However, it is possible that it is still too early to see a shift based on these global measures. Further work with older children is needed to determine if a shift in language proficiency is evident for bilingual children. Current efforts in our laboratory are focused on following a subset of these children until about 12 years of age. Also, it is possible that global measures of morphosyntax and grammaticality, in particular, do not explain these changes in language proficiency and that more fine-grained measures at the sentence level are needed. Lastly, grouping all errors using these global indexes might obscure developmental trends that would be observed otherwise at the sentence level or by type of error, for example.

Differences in MLUw, PGU, and SI in Bilingual Children With and Without DLD

The difference in performance observed at the onset in bilingual children with DLD in comparison to their TD bilingual children supports what has been reported in the literature for monolingual children (e.g., Rice, 2012, 2013). Children with DLD were outperformed by TD children on all measures examined in this study in both languages. Also, as previously reported for monolingual children with DLD, bilingual children with DLD exhibit strength in their language growth. In this study, bilingual children with DLD showed average gains that were the same (PGU) or larger (MLUw and SI) than those for TD children. Bilingual children with DLD followed the same patterns of growth as children without DLD. When growth was evident for TD children, children with DLD also showed growth. When a plateau or maintenance effect was observed for TD children, as in the case of PGU, the same plateau was observed for children with DLD. These findings suggest that bilingual children with DLD grow in their English skills and grow and/or maintain their Spanish skills in similar ways as TD children. These findings are in alignment with the “growth signaling disruption” hypothesis (Rice, 2012), which posits that DLD is the result of disruptions in the neurobiological timing mechanisms regulating the onset, acceleration, and deceleration of language mechanisms rooted in genetic and epigenetic factors. The growth signaling disruption hypothesis predicts that the shape of growth is preserved in children with DLD, but the timing is shifted, which is consistent with what was observed in this study. Although more evidence is needed to support this hypothesis, it is the most plausible explanation for the growth patterns observed in this study.

Limitations and Future Directions

There are several limitations of this study that are important to acknowledge. First, this is a small sample of 100 children who vary in age at the onset of the study, which might affect the precision of the estimates at the various ages. Second, our longitudinal data set includes three time points, which allowed us to estimate linear growth only. It is possible that the growth of these children is better estimated using nonlinear approaches (e.g., quadratic), which would require additional time points. Third, while we obtained parent-reported estimates of children's exposure to Spanish and English and documented the language of instruction at school, we did not include these as predictors due to our limited sample size. Future studies should examine these effects and include other sources of input, such as interactions with siblings, peers, teachers, and so forth. As mentioned earlier, the results of this study cannot be generalized to other contexts with different levels of bilingual support. Fourth, SI as a broad measure of syntactic complexity does not capture other types of syntactic complexity beyond amount of subordination. Fifth, future studies may look at fine-grained information at the utterance level to examine the relations between utterance length, grammaticality, and subordination, which may show different patterns from those observed in this study. In addition, future studies should use a more qualitative approach to describe the types of errors or the types of subordination used. Lastly, this study did not examine the role of socioeconomic status or parental level of education, which might be necessary to elucidate individual patterns of change in each language for bilingual children.

Clinical Implications

Measures derived from spontaneous language samples are often used in clinical practice to identify children with language disorders. The results of this study are of particular importance for clinical practice for three reasons. First, clinicians must be aware of the different growth patterns observed in bilingual children. Crucially, the plateaus of grammaticality in Spanish should not be interpreted as a sign of a language disorder since many TD children also exhibit this pattern. Second, the observed growth pattern in MLU and SI highlights the importance of utilizing spontaneous language measures in both languages during the assessment and monitoring process. Because both languages for these children are changing over time, it is crucial to understand how they are changing in context with support for both languages. Importantly, speech-language pathologists should consider the level of support for bilingualism when interpreting assessment results. For example, we would not expect to see as much growth in Spanish in a community with low support for the heritage language. Third, the potential for cross-linguistic influence in MLU and SI may suggest that interventions targeting utterance complexity (MLUw and SI) might yield results in both languages; however, this needs to be further explored in future research.

Conclusions

Spanish–English bilingual children with and without DLD showed growth in both Spanish and English over time for all measures examined in this study. Spanish grammaticality grew in these children, but gains were minimal in Spanish compared to English. Bilingual children with DLD showed strength in language growth, similar to TD children, but weaknesses in skills at the onset of the study for both languages. The results of this study must be interpreted with caution, taking into account the available support for Spanish provided by the context.

Data Availability Statement

The data set generated and/or analyzed during the current study is available from the corresponding author on reasonable request.

Funding Statement

Research reported in this publication was supported by National Institute on Deafness and Other Communication Disorders (NIDCD) Award Number K23DC015835 granted to Anny Castilla-Earls. Support for article preparation was also partially funded by NIDCD Award Number T32 DC013017 to Juliana Ronderos.