Variables associated with language attrition

Family language policy (including parental and familial language use) and attitudes toward the use of L1 and L2 are critical factors that foster or hamper HL development among DLLs (see Quay & Montanari, Reference Quay, Montanari, Nicoladis and Montanari2016, for a review). Families that primarily use the HL increase children’s opportunities for HL development and indirectly positively influence their perception of it. On the other hand, families who primarily use the societal language limit children’s access to the HL, adversely affecting its development and, possibly, children’s attitudes towards it. In fact, research has shown that L1 attrition occurs more rapidly when parents are bilingual – and thus speak the societal language – because, in this context, not enough communication in the HL may be provided for children to become active bilinguals and children can use the L2 and still be understood by parents (Quay & Chevalier, Reference Quay, Chevalier, Montanari and Quay2019).

Children’s early immersion into English-only preschool programs may also accelerate HL attrition. Wong-Fillmore (Reference Wong-Fillmore1991) surveyed over 1,100 immigrant families and found that children who attended English-only preschool programs were much more likely than children enrolled in bilingual programs to experience L1 attrition. Indeed, HL competence before children enter school is no guarantee of later bilingualism when schooling occurs early and exclusively in the majority language. On the contrary, since the vast majority of DLLs in the US attend schools in which English is the primary or sole language of instruction, an accelerated growth in English concomitant to a deceleration in the development of Spanish after entering the school system are the norm in this population (Hammer et al., Reference Hammer, Komaroff, Rodriguez, Lopez, Scarpino and Goldstein2012; Hoff et al., Reference Hoff, Quinn and Giguere2018; Montanari et al., Reference Montanari, Ochoa and Subrahmanyam2019, Reference Montanari, Mayr and Subrahmanyam2022; Rojas & Iglesias, Reference Rojas and Iglesias2013). This is because, as children enter school, contexts for speaking English increase, whereas those for using Spanish diminish.

Other demographic, social, and individual variables interact with parental language use and attitudes and early English immersion to affect the process of L1 maintenance in DLLs. For instance, several studies have found that gender predicts successful HL acquisition (Arriagada, Reference Arriagada2005; Portes & Hao, Reference Portes and Hao1998; Portes & Rumbaut, Reference Portes and Rumbaut2001; Rojas & Iglesias, Reference Rojas and Iglesias2013; Zentella, Reference Zentella1997). Zentella (Reference Zentella1997), in an ethnographic study of Puerto Ricans in New York, found that females tended to have higher levels of Spanish maintenance and proficiency than males from the same neighborhood, and Spanish language use was associated more with female than male domains. Portes and Hao (Reference Portes and Hao1998) also documented better performance on HL proficiency measures among second-generation females than males, a finding that was interpreted as resulting from girls spending more time with their parents and thus receiving more exposure to the HL. Rojas and Iglesias (Reference Rojas and Iglesias2013) further studied gender differences in HL development in a longitudinal study of 1,723 Spanish–English-speaking children aged 5;7 years at the beginning of their first three years of formal schooling. Boys and girls showed similar growth patterns in Spanish and English, although the shapes of their trajectories differed. In Spanish, girls had higher initial proficiency levels than boys, and their growth rates were consistently higher across different language measures throughout the study. In English, boys and girls started at similar levels, but girls showed faster growth in the spring semesters. Hammer et al. (Reference Hammer, Davidson, Lawrence and Miccio2009) did not find that gender predicted vocabulary and early literacy skills in either Spanish or English among Spanish-speaking preschoolers and kindergarteners. However, the study documented differences in language usage based on the gender of the child: mothers of daughters were eight times more likely to speak to them using more or all Spanish when compared to mothers of sons. Mothers of sons were twice as likely to speak to their sons using more or all English than the mothers of girls. Hammer et al. (Reference Hammer, Davidson, Lawrence and Miccio2009) speculated that these language use patterns may derive from traditional gender norms where girls, but not boys, are socialized into staying at home and becoming parents (hence being immersed in the home Spanish environment) (Zentella, Reference Zentella1997).

Research on Latinx families also suggests that parental levels of acculturation (i.e., orientation to American culture) and enculturation (i.e., orientation to Latinx culture) (Gonzales et al., Reference Gonzales, Knight, Birman, Sirolli, Maton, Schellenbach, Leadbeater and Solarz2004) relate to L1 use in the home and children’s HL learning. Typically, English acquisition and usage are evidence that an immigrant parent has acculturated to mainstream society, and their children may be more exposed to English than the HL. In contrast, the retention and more frequent use of the L1 by the immigrant parent may indicate lower levels of involvement and acculturation to the host country but increased HL learning in children (Phinney & Flores, Reference Phinney and Flores2002). Boyce et al. (Reference Boyce, Gillam, Innocenti, Cook and Ortiz2013) found a positive and significant correlation between maternal acculturation and total vocabulary size (including both English and Spanish words) in Spanish–English bilingual toddlers at 24 and 36 months of age. However, the study did not separately examine the influence of maternal acculturation and enculturation on children’s English and Spanish language outcomes, respectively. Cote and Bornstein (Reference Cote and Bornstein2014) studied bilingual American children, aged 20 months, born to Korean, Japanese, and South American immigrant mothers. They found that more acculturated mothers exposed their children more to English, leading to larger English vocabularies in the children. On the other hand, more enculturated immigrant mothers reported increased use of the HL with their children, resulting in larger HL vocabularies for the children.

Lexical and grammatical patterns of language attrition in Spanish–English dual language learners

Few studies have specifically looked at L1 attrition in young Spanish–English-speaking children. Early investigations were case studies of a few children (Anderson, Reference Anderson1999a, Reference Anderson1999b, Reference Anderson2001); other seminal studies focused on adults. For example, Montrul (Reference Montrul2008, Reference Montrul2016) documented extensive and often subtle differences in the lexical and morphosyntactic competence of adult Spanish HL speakers with respect to monolingual Spanish speakers. In the author’s words (Montrul, Reference Montrul2012, p. 4), by the time HL learners become adults, they display “non-native like competence and use of the language, better ability with receptive than productive language, non-uniform levels of proficiency, and linguistic gaps that resemble the patterns attested in second language acquisition (in gender agreement, verb paradigms, pronouns, case marking, word order, prepositions, etc.).” As a matter of fact, Montrul (Reference Montrul2016) points out that many of the differences between HL speakers and monolinguals may be due to differential acquisition of the L1 because these speakers may not have had the opportunity to fully acquire the morphosyntactic aspects of their L1. See Silva-Corvalán (Reference Silva-Corvalán2014, Reference Silva-Corvalán, Shin and Erker2018) for a similar argument.

Other studies that can inform us about L1 attrition have focused on bilingual development in DLLs, examining overall measures of language competence (often measured through parental reports) rather than specific lexical and syntactic patterns. For example, Hoff et al. (Reference Hoff, Quinn and Giguere2018), who studied the English and Spanish lexical and grammatical abilities of Spanish–English bilingual children between 30 and 48 months using parental reports, documented a negative relationship between level of English skills (either in vocabulary or grammar) and subsequent Spanish growth, suggesting that children with higher English proficiency early on were more likely to display lower Spanish competence with time. These results were interpreted as evidence that growing competence in the societal language threatens children’s continued acquisition of the HL. Hammer et al. (Reference Hammer, Lawrence and Miccio2008, Reference Hammer, Komaroff, Rodriguez, Lopez, Scarpino and Goldstein2012) also investigated the Spanish and English developmental trajectories of bilingual preschoolers from preschool entry at around 3;6 years of age to program exit at approximately age 5. Hammer et al. (Reference Hammer, Lawrence and Miccio2008) focused on receptive vocabulary and language comprehension abilities and found that children’s raw scores on the English receptive vocabulary test accelerated, while children’s standard scores on the Spanish language comprehension measure decelerated after an initial period of linear growth. Thus, despite Spanish exposure at home, the participants appeared to be losing their Spanish comprehension abilities after a significant change in the language environment occurred. Hammer et al. (Reference Hammer, Komaroff, Rodriguez, Lopez, Scarpino and Goldstein2012) found that simultaneous bilinguals began preschool with comparable productive vocabularies in the two languages but English vocabularies were larger than Spanish vocabularies by the end of preschool, and growth in English was strong, whereas it was weak in Spanish. On the other hand, sequential bilinguals, who predominantly began to acquire English in the preschool program, had larger Spanish vocabularies throughout preschool. However, there was stronger growth in English than in Spanish, suggesting, as in other studies, accelerated growth in English concomitant to a deceleration in the development of Spanish after entering preschool (Hammer et al., Reference Hammer, Komaroff, Rodriguez, Lopez, Scarpino and Goldstein2012; Hoff et al., Reference Hoff, Quinn and Giguere2018; Montanari et al., Reference Montanari, Ochoa and Subrahmanyam2019, Reference Montanari, Mayr and Subrahmanyam2022). It is important to point out, however, that for children who spoke predominantly Spanish at home, attending English-language preschool did not result in a dominant language shift and/or L1 attrition.

Research on how L1 attrition affects production abilities in developing Spanish–English DLLs is scarce and mostly limited to case studies. These studies have shown that the lexicon and the grammatical system are the most affected areas of development during this process (Anderson, Reference Anderson1999a, Reference Anderson1999b, Reference Anderson2001; Castilla-Earls et al., Reference Castilla-Earls, Francis, Iglesias and Davidson2019; Guiberson et al., Reference Guiberson, Barrett, Jancosek and Yoshinaga Itano2015; Hiebert & Rojas, Reference Hiebert and Rojas2021; Silva-Corvalán, Reference Silva-Corvalán2014, Reference Silva-Corvalán, Shin and Erker2018). Indeed, the reduction of L1 input and use lowers the speed and accuracy of lexical retrieval, eventually causing loss of lexical knowledge and a narrowing of the vocabulary (see Anderson, Reference Anderson and Goldstein2022, for a review). Early patterns of lexical loss include a reduction in the production of nouns, followed by a reduction in the production of verb lexemes (Anderson, Reference Anderson and Goldstein2022). For instance, in a longitudinal case study of a Spanish–English bilingual child who was experiencing L1 attrition, Anderson (Reference Anderson1999a) documented a significant decline in the use of different nouns and action verbs over time and showed an increase in the use of general terms such as the demonstrative pronouns éste, “this,” or ése, “that” (e.g., quiero ése, “I want that,” instead of quiero el juguete, “I want the toy”). Other studies have documented an increase in code-switching to English to make up for Spanish vocabulary gaps (Anderson, Reference Anderson1999a, Reference Anderson1999b, Reference Anderson2001; Guiberson et al., Reference Guiberson, Barrett, Jancosek and Yoshinaga Itano2015; Hiebert & Rojas, Reference Hiebert and Rojas2021; Montanari et al., Reference Montanari, Ochoa and Subrahmanyam2019). Relatedly, Hiebert and Rojas (Reference Hiebert and Rojas2021), who examined the trajectories of Spanish language growth and loss in 34 Spanish–English DLLs from preschool to kindergarten, found that lexical diversity in Spanish (as measured by the Moving-Average Type-Token Ratio, MATTR) suffered from a significant decline over time, especially when code-switched utterances were excluded from the analyses.

Reduction in input and output also has an impact on the grammatical skills of Spanish–English DLLs, resulting, in particular, in a progressive reduction of productivity and inflectional morphology with possible regularization of irregular forms, errors at the morphosyntax-semantics-pragmatics interface, including with determiners and subject pronouns, and the transfer of L2 syntactic structure to the L1 (Anderson, Reference Anderson1999a, Reference Anderson1999b, Reference Anderson2001; Silva-Corvalán, Reference Silva-Corvalán2014, Reference Silva-Corvalán, Shin and Erker2018; see also Montrul, Reference Montrul and Potowski2018, for a review). Hiebert and Rojas (Reference Hiebert and Rojas2021) specifically documented a significant deceleration of mean length of utterance in words (MLUw) in Spanish coupled with a decrease in the proportion of grammatical utterances (PGU) between preschool and kindergarten. Similarly, Castilla-Earls et al. (Reference Castilla-Earls, Francis, Iglesias and Davidson2019) reported a decrease in PGU in Spanish between kindergarten and second grade even in children receiving Spanish–English bilingual instruction. It has been reported that features of the noun phrase and verb morphology are particularly affected in HL development when children experience a shift from Spanish to English dominance (Castilla-Earls, Pérez-Leroux, et al., Reference Castilla-Earls, Pérez-Leroux, Martinez-Nieto, Restrepo and Barr2020; Montrul, Reference Montrul, Stavans and Jessner2022).

At the noun phrase level, mismatches in gender agreement and case are frequent (Anderson, Reference Anderson1999a, Reference Anderson1999b; Anderson & Márquez, Reference Anderson, Márquez and Grinstead2009; Hiebert & Rojas, Reference Hiebert and Rojas2021). Indeed, studies of heritage Spanish-speaking children in the US experiencing L1 attrition reveal that the main mismatches are in the use of a masculine article with a feminine noun or vice versa (e.g., *el mesa, “the (masc.) table (fem.),” rather than la mesa, “the (fem.) table (fem.),” or *la perro, “the (fem.) dog (masc.),” rather than el perro, “the (masc.) dog (masc.)”) or the use of a singular article for a plural one (*un perros, “a dogs”) (Castilla-Earls, Pérez-Leroux, et al., Reference Castilla-Earls, Pérez-Leroux, Martinez-Nieto, Restrepo and Barr2020; Martinez-Nieto & Restrepo, Reference Martinez-Nieto and Restrepo2023). At the verbal phrase level, changes affect the use of tense, aspect, mood, person, and number distinctions. Tense errors occur when a required tense (e.g., past tense as in se fue, “he left”) is replaced by another tense, for example, the simpler and more default present tense (se va, “he leaves”). Examples of aspect errors include the use of a perfective tense (such as the perfect tense in comí, “I ate”) when an imperfective tense is required (i.e., the imperfect tense in comía, “I used to eat”). Finally, examples of mood errors include the use of the indicative mood (e.g., no quiero que lo *sabe, “I don’t want that s/he knows”) for the subjunctive mood (e.g., no quiero que lo sepa, “I don’t want that he/she knows”) in instances in which the subjunctive form is the necessary one (Anderson, Reference Anderson2001; Silva-Corvalán, Reference Silva-Corvalán2014, Reference Silva-Corvalán, Shin and Erker2018; see also Montrul, Reference Montrul and Potowski2018, for a review). Person and number distinctions are also aspects of grammar that are adversely affected by L1 attrition after early L2 exposure, resulting in errors with subject-verb agreement. For example, Anderson (Reference Anderson2001) reported that singular forms replaced plural ones (e.g., duerme, “he/she sleeps” for duermen, “they sleep”), and the third person singular form was overextended to all other forms (quiere, “he/she wants,” for quieres, “you want”). Thus, the trend was a reduction of the person/number paradigm and the collapsing of all forms to a general, single one: the more default third person singular form. The regularization of irregular verb forms (i.e., sabo for sé, “I know”) has also been documented in the extant literature, although children’s patterns of overregularization have been shown to be inconsistent and were evidenced with some, but not all, irregular verb forms (Anderson, Reference Anderson2001; Silva-Corvalán, Reference Silva-Corvalán2014, Reference Silva-Corvalán, Shin and Erker2018), suggesting that factors such as complexity, saliency, and frequency of occurrence in the input and output may be responsible for this pattern.

The present study

The goal of this study is to examine the changes in the Spanish lexical and grammatical productive skills of 26 young Spanish–English DLLs from the beginning of their first year of preschool (average age 3;7) to a year later. Since previous studies on bilingual development in DLLs have focused on children’s language abilities as reported by parents (Hoff et al., Reference Hoff, Quinn and Giguere2018), on receptive vocabulary and language comprehension skills (Hammer et al., Reference Hammer, Lawrence and Miccio2008, Reference Hammer, Komaroff, Rodriguez, Lopez, Scarpino and Goldstein2012), or on a limited number of production variables in preschoolers (Hiebert & Rojas, Reference Hiebert and Rojas2021) and school-aged children (Castilla-Earls et al., Reference Castilla-Earls, Francis, Iglesias and Davidson2019), we examine the patterns of change in HL proficiency during the preschool years as measured by a wide range of lexical and grammatical measures in Spanish spontaneous speech, including 1) the number of complete and intelligible utterances (NCIU), 2) the total number of words (TNW), 3) the number of different words (NDW), 4) the mean length of utterance in words (MLUw), 5) the proportion of utterances with verbs (PUV), 6) the number of omitted words (NOW), and 7) the proportion of grammatical utterances (PGU).

Furthermore, since the process of language development and attrition is affected by a variety of external and internal factors that have been scarcely investigated (e.g., maternal cultural orientation) or poorly understood (e.g., gender), we run multiple linear models to investigate the extent to which demographic, social, and individual variables – in particular, age, gender, and mother’s enculturation and acculturation – predict each of the outcome proficiency variables. Finally, because the areas of language production that are most affected by L1 attrition are the grammatical system (Anderson, Reference Anderson and Goldstein2022), we conduct a qualitative analysis of children’s deviations from standard Spanish grammatical production between ages 3;7 and 4;7. We thus ask the following research questions:

1. 1. To what extent do children’s Spanish lexical (TNW, NDW) and grammatical (NCIU, MLUw, PUV, NOW, PGU) skills displayed in spontaneous speech change during the first year of English-only instruction between ages 3;7 and 4;7?

2. 2. Do children’s age, gender, and mothers’ enculturation and acculturation predict each of the outcome proficiency variables?

3. 3. To what extent do children’s production of articles, copula verbs, and prepositions at ages 3;7 and 4;7 deviate from standard Spanish grammatical forms?

Based on the extant literature (Castilla-Earls et al., Reference Castilla-Earls, Francis, Iglesias and Davidson2019; Hammer et al., Reference Hammer, Komaroff, Rodriguez, Lopez, Scarpino and Goldstein2012; Hiebert & Rojas, Reference Hiebert and Rojas2021; Hoff et al., Reference Hoff, Quinn and Giguere2018; Montanari et al., Reference Montanari, Ochoa and Subrahmanyam2019, Reference Montanari, Mayr and Subrahmanyam2022), we hypothesize that the children’s Spanish lexical and grammatical skills may show limited growth between ages 3;7 and 4;7, as children are introduced to formal English-only schooling. We also expect that features of the noun phrase and verb morphology will be particularly affected by increased English exposure and a reduction of Spanish input. We are unable to put forward a conclusive hypothesis as to the extent to which children’s age, gender, and mothers’ cultural orientation will predict children’s Spanish productive skills given that studies have produced mixed results as to the relevance of these factors.

Procedures

In order to measure mothers’ cultural orientation, mothers were administered the Acculturation Rating Scale for Mexican Americans-II (ARSMA-II, Cuéllar et al., Reference Cuéllar, Arnold and Maldonado1995), which independently assessed their orientation to Mexican (enculturation) and Anglo-American culture (acculturation). The scale consists of an Anglo orientation subscale (AOS) with 13 items and a Mexican orientation subscale (MOS) with 17 items that assess various aspects of cultural orientation, including language use and preferences, ethnic identity and classification, cultural heritage and ethnic behaviors, and ethnic interaction. Each question is scored on a Likert scale from 1 (“not at all”) to 5 (“extremely often or almost always”). Mean scores for each subscale are calculated by adding the scores of all items and dividing it by the total number of questions. Higher AOS and MOS scores represent higher levels of acculturation and enculturation, respectively, whereas lower scores represent less orientation to that specific culture. Reliability and test-retest reliability for ARSMA-II scales are high as indicated by -0.83 and 0.88 correlations for the AOS and MOS, respectively (Cuéllar et al., Reference Cuéllar, Arnold and Maldonado1995). Correlations between acculturation scores from the original ARSMA and those from the ARSMA-II have further revealed strong construct and concurrent validity as well as high convergent validity for the ARSMA-II, suggesting that it is a valid and reliable measure to assess acculturation among Mexican-Americans (Jimenez et al., Reference Jimenez, Gray, Cucciare, Kumbhani and Gallagher-Thompson2010).

Table 1 shows the mothers’ enculturation and acculturation scores. The maternal mean enculturation score was 4.48 (SD: 0.37; range: 3.18-5.00), suggesting that mothers were “very Mexican oriented,” a categorization that accurately reflected their status as recent immigrants in a region characterized by a large Mexican community. In contrast, their mean acculturation score was 2.29 (SD: 0.79; range: 1.08-4.23), which indicated a more limited orientation to American culture. Thus, as shown by the standard deviations and ranges, mothers differed substantially in how much they were oriented to American culture, whereas they were more uniform in their Mexican orientation.

In order to examine children’s Spanish lexical and grammatical skills, we collected naturalistic speech samples at preschool entry and a year later during which children interacted with a research assistant who, although bilingual, posed as a monolingual Spanish speaker. Analysis of spontaneous language, which allows for a naturalistic observation of a child’s representative language skills, is one of the standard strategies for the assessment of child language (Gutiérrez-Clellen et al., Reference Gutiérrez-Clellen, Restrepo, Bedore, Peña and Anderson2000; Simon-Cereijido & Gutiérrez-Clellen, Reference Simon-Cereijido and Gutiérrez-Clellen2007). While research with school-age children has particularly relied on narrative samples (Rojas & Iglesias, Reference Rojas and Iglesias2013), language samples produced during play have long been used to document, in an effective and ecologically valid way, the expressive language development of preschoolers with and without language disorders (Binger et al., Reference Binger, Ragsdale and Bustosa2016).

Before collecting the data, the research assistant spent time in the Head Start classroom in order to familiarize herself with the children. During the data collection, each child was individually taken to a quiet room and recorded for approximately 45 minutes while interacting with the research assistant around a set of age-appropriate toys and books, including a food set, a car set, a doll set, a farm play set as well as the books Frog Where Are You? (Mayer, Reference Mayer1969) and A Boy, a Dog, and a Frog (Mayer, Reference Mayer1967). Because the goal was to elicit spontaneous speech, the research assistant, who was trained in eliciting complex language, interacted with the children naturally, asking open-ended questions that focused on the toys/books at hand but also on the child’s interests and leads. While Spanish and English speech samples were collected at both times, the current study is based only on the Spanish samples. See Montanari et al. (Reference Montanari, Mayr and Subrahmanyam2018, Reference Montanari, Ochoa and Subrahmanyam2019, Reference Montanari, Mayr and Subrahmanyam2021, Reference Montanari, Mayr and Subrahmanyam2022) for further details on the data collection protocol and the children’s speech and language skills in both Spanish and English.

Transcription and coding

The speech samples were transcribed orthographically by two research assistants and reassessed by a third transcriber until consensus on all transcriptions was reached. Then each transcription was coded and analyzed using Systematic Analysis of Language Transcripts (Miller & Iglesias, Reference Miller and Iglesias2017). Since we were interested in examining the children’s grammatical abilities, we coded the first 100 multi-word utterances of each sample, excluding single words, for grammatical errors at the word and utterance levels. Traditionally, language samples with 50 to 100 utterances have been considered reliable for providing data that is representative of a child’s language skills (Miller et al., Reference Miller, Andriacchi and Nockerts2011; Paul & Norbury, Reference Paul and Norbury2012). As shown in the example (1) below, errors at the word level (EW) included errors with Spanish morphological production, omissions (marked with *), and overgeneralizations (EO), whereas errors at the utterance level (EU) mainly included word order errors.

A second transcriber performed inter-rater reliability of coding, and disagreements were discussed by re-examining the transcriptions until consensus was reached. We then used SALT to automatically generate the lexical and grammatical proficiency measures that have been previously used to evaluate oral language samples and identify language attrition (Anderson, Reference Anderson1999a, Reference Anderson1999b, Reference Anderson2001; Castilla-Earls et al., Reference Castilla-Earls, Francis, Iglesias and Davidson2019; Guiberson et al., Reference Guiberson, Barrett, Jancosek and Yoshinaga Itano2015; Hiebert & Rojas, Reference Hiebert and Rojas2021), including: 1) the number of complete and intelligible utterances (NCIU), 2) the total number of words (TNW), 3) the number of different words (NDW), 4) the mean length of utterance in words (MLUw), 5) the proportion of utterances with verbs (PUV), 6) the number of omitted words (NOW), and 7) the proportion of grammatical utterances (PGU). We excluded abandoned utterances, unintelligible segments, and utterances with code-switching from these analyses as described in previous research and the SALT manual (Gutiérrez-Clellen et al., Reference Gutiérrez-Clellen, Restrepo, Bedore, Peña and Anderson2000; Hiebert & Rojas, Reference Hiebert and Rojas2021; Miller & Iglesias, Reference Miller and Iglesias2017; Simon-Cereijido & Gutiérrez-Clellen, Reference Simon-Cereijido and Gutiérrez-Clellen2007).

Data analysis

Statistical comparisons of changes in children’s Spanish lexical and grammatical productive skills between 3;7 and 4;7 were completed using the Statistical Package for the Social Sciences (SPSS Statistics Version 27). Specifically, we ran mixed linear models (MLM) to investigate the extent to which 1) time of outcome measurement, 2) age at first measurement, 3) gender, 4) maternal enculturation, and 5) maternal acculturation predicted each of the outcome proficiency variables. MLM can compare change over time while allowing differences across different participants. The aforementioned five predictors were treated as the fixed effects, and time of measurement was treated as random effect. Because there were only two data points, only linear models were estimated for each outcome measurement.

We supplemented our quantitative analysis with a qualitative analysis of the nonstandard grammatical productions of the children from their entry into preschool to one year later. We specifically focused on identifying and extracting utterances that contained grammatical errors. We recorded the frequency of ungrammatical usage of articles, object clitic pronouns, copula verbs, verbs (excluding copula verbs), and prepositions for each individual child. In the few cases in which we could not determine the intended target words based on the transcript, we excluded the nonstandard productions from the qualitative analysis.

To compare the children’s nonstandard productions at two different time points, we employed paired t-test analyses. Additionally, we conducted a frequency analysis to provide a more detailed characterization of the grammatical errors. This analysis involved coding instances of omissions and mismatches in gender and number for articles. For copula verbs and prepositions, we coded instances of omissions and substitutions.

Number of complete and intelligible utterances (NCIU)

The only significant fixed effect predictor of the number of complete and intelligible utterances was time (NCIU, Figure 1). The coefficient of time was 10.19, which indicates a predicted increase of 10.19 complete and intelligible utterances from Time 1 to Time 2. Meanwhile, the intercept of NCIU was also significant, which means that there was a significant difference of baseline NCIU among the children. The random effect was not significant. The growth pace from Time 1 to Time 2 seemed to be the same for the children in this study.

Lexical measures

Children demonstrated growth in their lexical skills. Time was the only significant predictor of the total number of words (TNW, Figure 1). The coefficient was 84.69 and an increase of 85 words was observed from Time 1 to Time 2. Notably, the random effect of time was also found to be significant (p = 0.024). This indicates that the rate of change varied significantly among different participants, with individual children showing different growth patterns.

As to the number of different words (NDW, Figure 1), both time and gender were significant predictors. The coefficient of time was 27.38. Children demonstrated an increase of 27 different words from Time 1 to Time 2. Girls produced 14 different words more than boys. On average, girls produced 115 different words (SD = 21.46), while boys produced 100 (SD = 25.09). However, it is worth noticing that there was great variability within each gender group as evidenced by the standard deviations. Even though on average girls produced more different words, the inner group variability should be considered. Similarly, the random effect of time was also found to be significant (p = 0.034), suggesting significant different growth patterns among different participants.

Grammatical measures

Upon entering preschool, children demonstrated the ability to combine 3 to 4 words into phrases and sentences, albeit with some instances of missing words and grammatical errors, as expected based on the literature (Castilla-Earls & Eriks-Brophy, Reference Castilla-Earls and Eriks-Brophy2012; Simon-Cereijido & Gutiérrez-Clellen, Reference Simon-Cereijido and Gutiérrez-Clellen2007). Both time and gender predicted the mean length of utterance in words (MLUw, Figure 1). MLUw exhibited an increase of 0.57 words per utterance from Time 1 (M = 3.4, SD = 0.48) to Time 2 (M = 3.97, SD = 0.58). Thus, by Time 2, children produced utterances that included almost 4 words in length. Girls scored 0.36 points higher than boys in MLUw. Girls’ average MLUw was 3.89 (SD = 0.54), whereas boys’ average MLUw was 3.53 (SD = 0.66) at Time 1.

In terms of the proportion of utterances with verbs (PUV, Figure 1), time also emerged as a significant predictor. However, the increase in such utterances from Time 1 to Time 2 was relatively small, rising from 1% (SD = 0.01) to 2% (SD = 0.03).

As children progress in their language development, they are expected to omit fewer necessary words. The number of omitted words (NOW, Figure 1) was predicted, revealing time, again, as a significant negative predictor with a coefficient of -3.15. This indicates that, on average, children omitted 3 fewer words at Time 2 than Time 1. The average NOW at Time 1 was 6.96 (SD = 5.43), while at Time 2 it decreased to 3.81 (SD = 2.97).

Finally, the proportion of grammatical utterances (PGU, Figure 1) provided further insights into Spanish language development. Four significant predictors were identified in predicting PGU: time, gender, mothers’ enculturation and acculturation. PGU increased by 5% from Time 1 to Time 2, rising from 82% to 87%. Girls demonstrated a higher level of grammaticality, scoring 2% higher than boys. Interestingly, both mothers’ enculturation and acculturation positively influenced children’s grammaticality with the coefficient for enculturation to be 0.09 and the coefficient of acculturation to be 0.04. This means that when mothers exhibited a strong cultural orientation towards both Mexican and American cultures, children produced a greater number of grammatical utterances.

Qualitative analysis of nonstandard grammatical productions

Two grammatical measures, namely the number of omitted words and the proportion of grammatical utterances, indicate that children exhibited a range of ungrammatical productions at both time points. Table 4 provides the means and standard deviations for specific nonstandard instances of articles, clitic pronouns, copula verbs, other verbs, and prepositions in individual samples at each time point. These nonstandard productions encompassed both omission and commission errors. Paired t-tests suggested that children significantly improved grammaticality at Time 2 with the production of articles (t(25)=3.239, p =.003, d =.63) and copula verbs (t(25)=2.038, p =.052, d =.40). However, no significant differences were observed in the number of errors with object clitic pronouns, other verbs, and prepositions between the two timepoints.

Table 4. Means and Standard Deviations of Nonstandard Grammatical Productions in Each Sample at Time 1 and Time 2

Table 5 presents the frequency of nonstandard productions of articles, copula verbs, and prepositions across the entire sample at each time point. At Time 1, omission errors were more prevalent in copula verbs and prepositions. At the noun phrase level, gender mismatches in articles occurred more frequently compared to omissions or number mismatches. After one year, the sample exhibited an increase in substitution errors for copula verbs and a persistent prevalence of omission errors over substitutions for prepositions. Regarding articles, gender mismatches continued to be more common than number mismatches and omissions. Examples of children’s nonstandard productions can be found in Table 6.

Table 5. Frequency of Nonstandard Article, Copula Verb, and Preposition Productions in the Complete Sample at Each Time Point

Table 6. Examples of Nonstandard Productions

Changes in children’s spanish lexical and grammatical skills between 3;7 and 4;7

Unlike studies that have documented an apparent loss of Spanish lexical and grammatical skills as children enter preschool or elementary school (Castilla-Earls, Auza, et al., Reference Castilla-Earls, Auza, Pérez-Leroux, Fulcher-Rood and Barr2020; Hiebert & Rojas, Reference Hiebert and Rojas2021; Morgan et al., Reference Morgan, Restrepo and Auza2013), our study shows that entering the school system and receiving English-language instruction does not result in an abrupt dominant language shift or Spanish attrition, at least over one year and in a context characterized by substantial numbers of Spanish-speaking immigrants such as Los Angeles. We must acknowledge that our results may stem from our research context, and hence may not be generalizable to other Spanish–English bilingual children in the US who may not have access to Spanish through a variety of sources in their community.

In terms of language output, children became significantly more productive in Spanish at Time 2, both at the utterance and word level, regardless of age, gender, and mothers’ cultural orientation. Specifically, on average, children produced 10 more complete and intelligible utterances and employed 27 more different words within a set of 100 multiword utterances between Time 1 and Time 2, despite one year of English-only instruction. To further explore the nature of utterance output, after completion of the analyses, we calculated the average percentage of intelligible utterances at each time point. On average, 88% of the utterances at Time 1 were deemed intelligible, while this percentage increased to 91% at Time 2, indicating some expressive growth. Additionally, we examined the occurrence of code-switching within the set of 100 utterances. Children exhibited a similar number of utterances with code-switching at both time points, with an average of 14 utterances at Time 1 and 13.4 at Time 2. These findings suggest that the observed growth in productivity is primarily associated with enhanced intelligibility, as a comparable number of utterances with code-switching were excluded from the analysis set at both time points.

It is important to acknowledge that our study had an observation period of one year, unlike previous studies that have examined language skills over a longer duration. This limited timeframe may not have been sufficient to observe attrition in language proficiency. Prior research investigating Spanish–English DLLs entering preschool has often demonstrated a deceleration in the development of Spanish skills concurrent with accelerated growth in English skills, with studies following children for more than a year documenting an initial linear growth in Spanish lexical and syntactic measures (Hammer et al., Reference Hammer, Lawrence and Miccio2008, Reference Hammer, Komaroff, Rodriguez, Lopez, Scarpino and Goldstein2012; Hoff et al., Reference Hoff, Quinn and Giguere2018). Hence, it is crucial to consider the likelihood that as children progress in an English-speaking educational environment, there may be a decline in the pace of their Spanish development. Indeed, research focusing on Spanish–English-speaking children aged six and older consistently demonstrates a deceleration and attrition in their Spanish proficiency.

One crucial observation is that the Spanish productions of the children at Time 2 were still far from reaching full development. There are few published longitudinal studies of Spanish monolingual preschoolers’ spontaneous speech samples using a comparable methodology (Fernández Vázquez & Aguado Alonso, Reference Fernández Vázquez and Aguado Alonso2007). Yet, the progress in language production attested for our participants aligns with expected developmental patterns found in studies of Spanish monolingual children (Auza Benavides & Chávez Luján, Reference Auza Benavides, Chávez Luján, Benavides and Zimmermann2013; Camargo-Mendoza & Garayzábal-Heinze, Reference Camargo-Mendoza and Garayzábal-Heinze2015). For instance, in a study of morphosyntactic development of two groups of 33-to-52-month-old children learning two dialectical varieties of Spanish (21 children learning Mexican Spanish and 12 learning Castilian Spanish), language samples collected during play reveal an MLUw range between 3.06 to 5.5; however, only 36% of the children had MLUw from 3;06 to 3;90, as observed in this study, while the rest of the participants produced longer utterances (Johnson, Reference Johnson and Pérez Pereira1996). Another study of Colombian children between 10 months and 6;7 years of age examined MLUw in spontaneous speech across different age intervals (Camargo-Mendoza & Garayzábal-Heinze, Reference Camargo-Mendoza and Garayzábal-Heinze2015). The mean MLUw of the 11 children who were between ages 3;7 and 4;6 was 3.64 (SD = 0.36), comparable to our findings. A study of narratives produced by children from Mexico City also reports values for 24 four-year-old children (Auza Benavides & Chávez Luján, Reference Auza Benavides, Chávez Luján, Benavides and Zimmermann2013). The MLUw mean was 4.75 with an SD of 1.80, with a range from 0.63 to 7.74 words. However, MLUw are known to be longer in narration than in play-based samples. This study also evaluated NDW (mean = 66.86, SD = 24.54, range = 16-123), which was lower than what we found in this study. Similarly, a study of spontaneous measures based on narrations of Colombian children found that for 37 three-year-old and for 39 four-year-old children, the means of MLUw were 4.76 (SD = 1.23) and 5.96 (SD = 1.11), respectively (Castilla-Earls & Eriks-Brophy, Reference Castilla-Earls and Eriks-Brophy2012). Of note, the reported ranges for the two age groups were quite large: for the age 3 group, MLUw ranged from 2.00 to 7.41, and for the age 4 group, it ranged from 4.15 to 8.08. Furthermore, in a longitudinal study of 50 Spanish young children, MLU measures collected at age 3, 3;6, and 4 showed a significant increase every 6 months. However, on average, the annual increase was smaller than one word and was considered relatively small (Fernández Vázquez & Aguado Alonso, Reference Fernández Vázquez and Aguado Alonso2007). Overall, then, our results align with those found in studies of monolingual Spanish-speaking children.

Although the children in our study produced longer utterances and more verbs compared to Time 1, the percentage of utterances with verbs remained low at both time points. Specifically, the percentage was 1% at Time 1 and only 2% at Time 2, indicating that the children primarily generated phrases or sentence fragments during their interactions with the research assistants. The low percentages of utterances with verbs suggest that the children rarely produced full sentences. Johnson (Reference Johnson and Pérez Pereira1996) also found that young Spanish-speaking children produced a reduced number of different verbs during play (e.g., 232 different verbs among 3067 total utterances), a result that resonates with our findings. It is also worth considering that the data collection context, which involved play sessions with natural interaction between the research assistant and the children, may have impacted the children’s use of complete clauses. While this context provides ecological validity and reflects real-world language use, it might not have strongly encouraged the production of full clauses.

Furthermore, it is important to note that the slopes for the total number of words and the number of different words were significant, implying that different children exhibited varying development rates. Due to the limited sample size, we cannot pinpoint specific factors contributing to this individual variation. However, it is plausible that similar to previous studies (Rojas & Iglesias, Reference Rojas and Iglesias2013), the initial levels of lexical and grammatical skills at Time 1 may have shaped the differential developmental rates observed among the children.

Our analysis of the changes in the proportion of grammatical utterances and the number of omitted words over time provides additional evidence that language productions at Time 2 were still undergoing development. Throughout the study, there was an improvement in grammaticality, characterized by a significant increase in the production of grammatical utterances (from 82% to 87% of the samples) and a decrease in the number of omitted words (from approximately 7 to 4 words per sample). Notably, the children in our study exhibited higher percentages of grammaticality in their speech samples compared to previous studies with preschoolers (Hiebert & Rojas, Reference Hiebert and Rojas2021), although differences in methodologies could contribute to some of these variations. In fact, Castilla Earls and Eriks-Brophy (Reference Castilla-Earls and Eriks-Brophy2012) found that monolingual Spanish-speaking three- and four-year-olds were ungrammatical in 22% and 14% of the sample on average, respectively.

Nevertheless, it is important to highlight the persistent presence of nonstandard productions previously documented in the literature on language attrition after one year of preschool education in English. Specifically, while errors in articles and copula verbs demonstrated a significant decrease over time, errors in verbs, object clitic pronouns, and prepositions remained consistent throughout the study. Furthermore, we observed ongoing difficulties with noun phrase structures, particularly in the use of articles, where gender mismatches were more prevalent compared to number substitutions and omissions. These findings are consistent with previous research that has reported similar patterns. For example, a study by Morgan et al. (Reference Morgan, Restrepo and Auza2013) with older children also noted a higher incidence of gender substitutions (8%) in articles compared to number omissions (3%).

Monolingual Spanish-speaking children typically acquire basic grammatical gender, particularly determiner/noun agreement, by age three (Anderson, Reference Anderson1999b; Montrul & Potowski, Reference Montrul and Potowski2007). In Spanish, gender marking is generally reliable, as most nouns explicitly indicate their gender by canonical -a/-o word markers (Pérez-Leroux et al., Reference Pérez-Leroux, López, Barreto, Cuza, Marinescu, Yang and Colantoni2023). Monolingual Spanish children demonstrate early mastery of gender use in articles, with two-year-olds achieving high accuracy rates of 97-100% in naturalistic speech (Snyder et al., Reference Snyder, Senghas and Inman2001). However, our findings suggest that reduced exposure to Spanish leads to prolonged development in the acquisition of grammatical gender among Spanish–English-speaking children, which aligns with previous research (Anderson, Reference Anderson1999a, Reference Anderson1999b; Anderson & Márquez, Reference Anderson, Márquez and Grinstead2009; Hiebert & Rojas, Reference Hiebert and Rojas2021; Martinez-Nieto & Restrepo, Reference Martinez-Nieto and Restrepo2023). For instance, studies involving older bilingual children and using elicited tasks have identified delays in the production of articles in Spanish (Morgan et al., Reference Morgan, Restrepo and Auza2013), and children’s English proficiency levels have been found to predict the accuracy of Spanish articles (Castilla-Earls, Pérez-Leroux, et al., Reference Castilla-Earls, Pérez-Leroux, Martinez-Nieto, Restrepo and Barr2020). In elicitation studies, children often exhibit high performance but tend to favor masculine assignment (Martinez-Nieto & Restrepo, Reference Martinez-Nieto and Restrepo2023; Pérez-Pereira, Reference Pérez-Pereira1991).

In this longitudinal study, the participants were younger and in earlier stages of English learning compared to the bilingual participants in previous studies that have reported gender agreement mismatches. We conducted a comprehensive analysis to gain a deeper understanding of the nonstandard productions within our sample. Across both time points, our findings revealed that 58% of the gender mismatches in articles involved incorrect female assignments (using a feminine article with a masculine noun), 35% involved incorrect male assignments (using a masculine article with a feminine noun), and 7% of the responses were unscorable (e.g., the use of the neutral article “lo”). Specifically, at Time 1, 64% of the article gender mismatches corresponded to incorrect female assignments (to masculine nouns), and this percentage decreased to 50% at Time 2. Interestingly, this trend contradicts previous studies that frequently reported male assignments to female nouns, positing the male gender as the default noun gender in Spanish (Goebel-Mahrle & Shin, Reference Goebel-Mahrle and Shin2020). However, it is worth noting that not all studies, especially those analyzing naturalistic language samples, have observed this pattern (Pérez-Leroux et al., Reference Pérez-Leroux, López, Barreto, Cuza, Marinescu, Yang and Colantoni2023). Moreover, early studies of Spanish development revealed an overgeneralization of feminine gender to masculine nouns (Hernández-Pina, Reference Hernández-Pina1984; Pérez-Pereira, Reference Pérez-Pereira1991). Regarding number mismatches in articles, they occurred less frequently than gender substitutions. In 69% of the cases, children did not apply plural marking to the articles, resulting in errors of number agreement.

Regarding nonstandard substitutions for copula verbs, approximately 60% of the errors involved using singular forms instead of plural forms, while the reverse (plural for singular) was not observed. These findings are in line with the results of Spanish monolingual and bilingual development research, which has also shown that Spanish L1 learners (Radford & Ploennig-Pacheco, Reference Radford and Ploennig-Pacheco1995), Spanish–Euskera bilingual children (Ezeizabarrena, Reference Ezeizabarrena, Pérez-Leroux and Glass1997), and Spanish–English bilingual children (Silva-Corvalán & Montanari, Reference Silva-Corvalán and Montanari2008) acquire singular forms before plural forms and substitute plural verb forms with singular ones in the early stages of development. Tense and person substitutions were infrequent, with only one instance of each. Incorrect assignment of ser/estar was also rare, with only five observations (three instances of using estar instead of ser and two instances of using ser instead of estar).