Introduction
Multilingualism is a typical aspect of everyday life for much of the world's population. While no statistics exist as to the number of people who function with multiple languages on a daily basis, the fact that the world's almost 7,000 languages are spoken in approximately 200 countries suggests that multilingualism is a characteristic of practically every region (Romaine, Reference Romaine, Bhatia and Ritchie2013). For instance, despite its reputation as a monolingual English-only country, at least 364 languages are spoken in the United States (Mackey, Reference Mackey, Bhatia and Ritchie2013), and cities like New York City, Los Angeles, and Chicago are largely multilingual. Second only to English, Spanish is the language most used in the country, and, according to US 2018 Census estimates, it is spoken by approximately 40 million Hispanic individuals (www.census.gov). In these families, Spanish may be the primary home language; yet, children hear both Spanish and English from early on and are formally and consistently exposed to English in preschool or kindergarten, while their Spanish is still being developed. The number of such children – DUAL LANGUAGE LEARNERS (Paradis, Genesee, & Crago, Reference Paradis, Genesee and Crago2011) – in US schools has doubled in recent years and Spanish-speaking children represented 77% of all English learners in 2015 (National Center for Education Statistics, 2017). These statistics suggest that bilingual children make up a large proportion of students in the typical classroom and the use of two languages – including in the same sentence – is common as children are educated in the majority language.
Despite the prevalence of what are uniformly referred to as dual language learners, children from bilingual backgrounds can be very different from one another since they experience a tremendous range of variability in their language histories, in particular, their onset of acquisition of, the amounts of exposure to, and the kinds of experiences and conversational partners in their two languages (Hoff & Core, Reference Hoff and Core2013). Moreover, given the dynamic nature of bilingualism, their patterns of language exposure and use are likely to fluctuate over time, with major changes occurring after entering daycare, preschool, or kindergarten. The beginning of schooling may also mark the first time that children are faced with monolingual English speakers and with monolingual patterns of language use in the majority culture, thus becoming aware of the sociolinguistic characteristics of their environment and the status of their languages (Gutiérrez-Clellen, Simón-Cereijido, & Leone, Reference Gutiérrez-Clellen, Simón-Cereijido and Leone2009). Indeed, it is often during this time that young bilinguals learn that English is the majority language, the language with the highest status in the community, and the preferred language in the school context, while Spanish is a minority language spoken only at home and by a segment of the population.
In this subtractive context, the practice of alternating between two languages in a single utterance or conversation – CODE or LANGUAGE MIXING (Meisel, Reference Meisel1994) – may raise particular concerns among teachers and service providers since, being non-existent in monolingual speech, it may be misinterpreted as a sign of confusion or of lack of proficiency or communicative competence (Hughes, Shaunessy, Brice, Ratliff, & McHatton, Reference Hughes, Shaunessy, Brice, Ratliff and McHatton2006). Echoes of this view can also be found in contemporary research. Some recent studies conducted in a variety of settings across the world have indeed shown that language mixing in young children is the result of proficiency-related factors, namely, lexical gaps that children fill in by borrowing words from their dominant language when interacting in their weaker language (see Genesee & Nicoladis, Reference Genesee, Nicoladis, Hoff and Shatz2007, for a review). At the other extreme, however, it has also been suggested that language alternation in discourse by children is an indicator of bilingual competence – so that more proficient bilinguals mix languages more than less proficient ones – because it conforms to the linguistic and pragmatic constraints of the two languages (Paradis, Nicoladis, & Genesee, Reference Paradis, Nicoladis and Genesee2000; Yow, Patrycia, & Flynn, Reference Yow, Patrycia, Flynn, Nicoladis and Montanari2016). Researchers have also found that children's patterns of language mixing may be also mediated by factors such as the interlocutor/community's language preferences, language status and prestige, setting, and topic (Gutiérrez-Clellen et al., Reference Gutiérrez-Clellen, Simón-Cereijido and Leone2009; Paradis & Nicoladis, Reference Paradis and Nicoladis2007). Furthermore, since dual language learners are a heterogeneous linguistic group whose language proficiency and use change over time (Hoff & Core, Reference Hoff and Core2013), their patterns of language mixing may be dynamic and characterized by variability – which measures of average such as means may fail to capture – and there may be different profiles of children with distinct proficiency and language mixing patterns.
It is the goal of this paper to examine language choice in bilingual preschoolers longitudinally, taking proficiency, variability, and sociolinguistic factors into account. In particular, we investigate patterns of language mixing in a sample of 26 Spanish–English bilingual children from Los Angeles, California, examining how these changed from the beginning of their first year in preschool to a year later when the children interacted in monolingual English and Spanish contexts. Since language mixing at early stages of development has been often interpreted as caused by vocabulary gaps, we examine whether the children's language mixing is related to their proficiency in Spanish and English. In addition, given the variability that characterizes dual language learners’ language histories, acquisition, and performance, we conduct an exploratory cluster analysis to assess whether there were different groups of dual language learners sharing similar proficiency and language use characteristics. We also take sociolinguistic variables into account to explain language mixing patterns that occur while interacting in the majority language, English, as opposed to those that occur when children use the minority language, Spanish.
Adults’ code-switching and children's code/language mixing
Although there is as yet no consensus concerning the use of terminology, switching between two languages in conversation is often referred to as CODE-SWITCHING (Poplack, Reference Poplack1980) when the juxtaposition of the two codes (languages or dialects) is grammatical and constrained by regularities, as in the case of proficient bilingual adults. However, researchers tend to use the term CODE or LANGUAGE MIXING to refer to young learners’ speech during the early stages of bilingual development, when it is debated whether such language alternation is as grammatically constrained as in the case of proficient bilingual adults (Meisel, Reference Meisel1994). Indeed, researchers have proposed that alternating between languages in conversation is different for individuals who have already established proficiency in two languages and young children who are still developing them. For instance, code-switching among bilingual adults tends to be socially motivated, especially in communities where it is typically practiced, and it serves to establish bilingual identity, tell a narrative, or include/exclude a particular individual from a conversation (Paradis et al., Reference Paradis, Genesee and Crago2011). Young bilingual children, on the other hand, often engage in language mixing due to vocabulary gaps (see Genesee & Nicoladis, Reference Genesee, Nicoladis, Hoff and Shatz2007, for a review); it is only after they become increasingly proficient in their two languages and are socialized into the preferred patterns of language use in their communities that their language mixing develops into the skill of code-switching (Jisa, Reference Jisa2000). In the present study, we will refer to the alternation between two languages as ‘code-switching’ among bilingual adults and as ‘language mixing’ among young dual language learners. We will also use the term ‘dual language learners’ and ‘bilingual children’ interchangeably to refer to children who are learning two languages in childhood.
A wealth of literature exists on code-switching in bilingual adults and children concerning the grammatical constraints evident in their speech (e.g., Bernardini & Schlyter, Reference Bernardini and Schlyter2004; Cantone, Reference Cantone2007; MacSwan, Reference MacSwan, Nicoladis and Montanari2016; Meisel, Reference Meisel1994; Poplack, Reference Poplack1980), as well as the social and pragmatic functions served by such utterances (e.g., Gumperz, Reference Gumperz1971; McClure, Reference McClure and Duran1981; Valdes-Fallis, Reference Valdes-Fallis1978). However, studies on language mixing among young bilingual learners that do not focus on grammar are fewer, and, until recently, they have been based primarily on case studies of individual children from different language communities (e.g., Genesee, Boivin, & Nicoladis, Reference Genesee, Boivin and Nicoladis1996; Genesee, Nicoladis, & Paradis, Reference Genesee, Nicoladis and Paradis1995; Paradis & Nicoladis, Reference Paradis and Nicoladis2007). At the same time, as pointed out above, different interpretations of this phenomenon have been proposed and there is as yet no consensus as to the reasons why adults and children mix languages. Some researchers argue that language alternation is an indicator of bilingual proficiency since it is a sophisticated and rule-governed behavior that is grammatically constrained (Poplack, Reference Poplack1980); it occurs in places within the utterance that do not violate the syntactic structure of either language (Poplack, Reference Poplack1980); and only individuals who are fully proficient in two languages are able to unconsciously comply with these norms (Meisel, Reference Meisel, Bathia and Ritchie2006).
Poplack (Reference Poplack1980), who analyzed the code-switching patterns of 20 Puerto Rican Spanish–English bilingual adults with different degrees of bilingual proficiency in New York, differentiated two types of language alternation – INTRA-SENTENTIAL and INTER-SENTENTIAL – arguing that they are indicative of different degrees of bilingual proficiency. Intra-sentential code-switching, which involves switching languages within the utterance, requires the most bilingual proficiency because the individual is forced to keep both language structures in mind while switching without violating the syntactic structure of either language. Inter-sentential code-switching, on the other hand, requires minimal bilingual proficiency because it involves switching between complete utterances. Since both language structures are separated at the utterance boundary, the individual does not have to engage in the difficult task of juxtaposing the two language structures before making the switch (Poplack, Reference Poplack1980). Findings from this study showed that participants who were competent in both languages engaged in a higher incidence of intra-sentential code-switching than Spanish-dominant speakers, who instead preferred the less risky inter-sentential code-switching. Peynircioglu and Durgunoglu (Reference Peynircioglu, Durgunoglu, Hereida and Altarriba2002) found similar results in a study of 36 Spanish–English bilingual three- to five-year-old preschoolers from Washington, DC, who had different proficiency levels in Spanish and English. The study revealed that, although there was an overall higher incidence of inter-sentential language mixing, the most balanced children engaged in intra-sentential language mixing more often than the less proficient dual language learners. Similar findings were reported in a completely different setting by Yow et al. (Reference Yow, Patrycia, Flynn, Nicoladis and Montanari2016), who studied 36 English–Mandarin bilingual kindergarteners and first-graders from Singapore. The bilingual children who mixed languages more – and at the intra-utterance level – indeed demonstrated higher proficiency in both Mandarin and English compared to those who mixed languages less, providing additional evidence that language mixing, particularly at the intra-sentential level, may be a mark of bilingual competence even at a young age.
However, other researchers working in similar communities have suggested that individuals mix languages for lexical gaps in one or both languages, and thus, language alternation – and in particular language mixing in young language learners – may be due to limited language proficiency (see Genesee & Nicoladis, Reference Genesee, Nicoladis, Hoff and Shatz2007, for a review) or even language loss (Anderson, Reference Anderson and Goldstein2012). This proposal, which predicts that individuals alternate languages more in contexts in which they are required to use their weaker language, can explain the language mixing patterns of children who are still in the process of developing two languages and thus resort to language mixing to fill in vocabulary gaps. Many case studies of individual children across different regions of the world have found, for instance, that young bilinguals borrow words or even syntactic structures from their dominant language when speaking their less developed language because they have not yet developed such translation equivalents or constructions in the weaker language (Bernardini & Schlyter, Reference Bernardini and Schlyter2004; Nicoladis & Secco, Reference Nicoladis and Secco2000). More recent investigations that include larger samples of children have also found that language mixing in children is proficiency-related. For instance, Greene, Peña, and Bedore (Reference Greene, Peña and Bedore2012), who examined single-word language mixes in a semantic task by 606 Spanish–English bilingual five-year-olds from Texas and Utah, reported that language dominance significantly impacted language choice as children tended to choose the language they knew better to name semantic items. Similarly, Ribot and Hoff (Reference Ribot and Hoff2014), who studied language choice in 115 two-and-a-half-year-old Spanish–English bilingual toddlers in South Florida using parental reports, found that the children who switched to English when interacting in Spanish (as reported by parents) showed greater English skills, specifically in the expressive domain. Children who were reported not to mix languages were more balanced bilinguals in both expressive and receptive skills. Although these recent large-scale studies did not examine language mixing in naturalistic conversations, their findings confirm the hypothesis that language mixing in young bilinguals is often the result of limited proficiency.
It must be noticed, however, that language mixing for vocabulary gaps by young bilinguals has not always been interpreted negatively; rather, some researchers (working in the same communities where the studies reviewed above were conducted) have seen it as a sign of communicative competence (Reyes, Reference Reyes2004). For instance, since dual language learners are still developing both languages, it is normal that they evidence lexical gaps (Genesee et al., Reference Genesee, Nicoladis and Paradis1995; Nicoladis & Secco, Reference Nicoladis and Secco2000). Furthermore, most dual language learners tend to be exposed to each language at different amounts and to situations in which the adults around them code-switch frequently. Given that children tend to reflect the language environment they are exposed to, it is of no surprise that these children engage in language mixing too. Instances of language mixing should thus not be seen as a sign of lack of proficiency at this age but as evidence of children's ability to resort to both of their languages effortlessly to continue the flow of the conversation and match their patterns of language use to those of their interlocutors or community (Hereida & Altarriba, Reference Hereida and Altarriba2001; Reyes, Reference Reyes2004).
For example, Comeau, Genesee, and Lapaquette (Reference Comeau, Genesee and Lapaquette2003) found that two- to three-year-old French–English bilingual children from Montreal were capable of varying their patterns of language choice following the lead of unfamiliar interlocutors who changed their rates of language mixing over three sessions. This finding was interpreted as evidence that young bilinguals can monitor language input on-line and adjust their language use accordingly, showing sensitivity to the interlocutors’ language skills, use, and preferences and thus advanced communicative competence. Yow et al. (Reference Yow, Patrycia, Flynn, Nicoladis and Montanari2016) also reported that the five-year-olds who came from a school in Singapore where Mandarin/English mixing was the norm mixed languages at a significantly higher rate (32% of the time) than proficiency-matched children whose school discouraged code-switching (1.49%), suggesting that children vary their patterns of language use following those in their environment. Reyes (Reference Reyes2004), examining the language mixing patterns of 20 seven- and ten-year-old Spanish–English bilinguals in a bilingual program in northern California, further found that the children who mixed languages the most – the older group – were more skilled in using both languages to keep the conversation flowing, thereby showing the greatest level of communicative competence. Thus, language mixing in children represents a controversial topic in the bilingual literature: while, at one extreme, some researchers have argued that it is the result of uneven bilingual proficiency, at the other extreme others have seen it as indicative of bilingual competence.
Models of code-switching that have been proposed to account for adults’ language alternation have posited that contextual factors such as interlocutor effects and speaker ecology are critical in explaining adult code-switching behavior. For instance, in the Control Process Model of code-switching (Green, Reference Green2018; Green & Wei, Reference Green and Wei2014), the environment – and the interactional context – will dictate to a large extent what the bilingual can and will do with regard to code-switching. For example, a single language context (common in stable bilingual communities with a tradition of language separation) or a dual language context where a speaker may converse in one language to one person and in another language to a different interlocutor (where there may be considerable asymmetry in speakers’ proficiency in the two languages) will result in single switches, in particular INSERTIONS and ALTERNATIONS (Muysken, Reference Muysken2000), which require competitive language control (Green, Reference Green2018) (in which items and constructions from only one language can be used). However, contexts in which languages have roughly equal prestige and no tradition of overt language separation (Muysken, Reference Muysken2000) or conversational partners who prime copious language mixing (Green, Reference Green2018) will elicit dense code-switching, which require cooperative language control, allowing items and constructions from either language into the utterance plan (Green, Reference Green2018).
While this model was not proposed with young developing bilinguals in mind, it may be used to explain some of the findings of other studies of children's language mixing. Some researchers have indeed argued that children's proficiency and sensitivity to their interlocutor/community's language patterns may interact with variables such as language status and prestige, setting, and topic, further affecting children's patterns of language choice. For example, in subtractive environments, arguably like the US, children are typically exposed to English rather than Spanish in the academic setting. Hence, many children may associate the school setting and school-related topics with the English language. Gutiérrez-Clellen et al. (Reference Gutiérrez-Clellen, Simón-Cereijido and Leone2009) conducted a study of language mixing among 58 six-year-old Spanish–English dual language learners with and without language impairment in Southern California. The children came from homes where Spanish was spoken predominantly; however, at school English was the language that was primarily used for instruction. The results showed that while English-dominant, six-year-old children tended to mix languages at a higher incidence when conversing in their non-dominant Spanish language, Spanish-dominant children tended to refrain from language mixing when speaking their non-dominant language, English. This was true both for typically developing children and for children with language impairment. The authors concluded that the Spanish-dominant children refrained from language mixing when speaking English, their weaker language, because, in the US, English tends to hold a higher status than minority languages, especially in the school setting (Gutiérrez-Clellen et al., Reference Gutiérrez-Clellen, Simón-Cereijido and Leone2009). For this same reason, the English-dominant dual language learners found it acceptable to mix more when speaking Spanish. Yip and Matthews (Reference Yip and Matthews2006) found similar results in a sample of six Cantonese–English bilinguals in Hong Kong. These children inserted English words into their Cantonese speech more frequently than vice versa, in line with the frequent code-switching behavior of the larger community in Hong Kong. These results point to the complexity of language mixing in young bilinguals and the importance of the sociolinguistic environment in accounting for this linguistic behavior.
The present study
It is the goal of this study to examine the language mixing patterns of 26 Spanish–English dual language learners in monolingual English and Spanish contexts longitudinally, from the beginning of their first year of preschool to a year later. Most investigations on this topic have often involved detailed case studies of eight or fewer children, or investigations with relatively large samples, but with no longitudinal examination of language mixing in naturalistic conversations. To our knowledge, this study is the first that attempts to bridge these two approaches by analyzing language mixing patterns in naturalistic conversations in a relatively large sample of children, and over the course of one year while they were being acculturated to the sociolinguistic characteristics of their environment. Since the participants were from predominantly Spanish-speaking homes, we were interested in observing their language mixing behavior after formal and consistent exposure to English began in preschool. We were particularly interested in examining (1) the incidence of intra-sentential versus inter-sentential language mixing at preschool entry (at 3;6) and a year later (at 4;5), and (2) whether these patterns were related to proficiency as measured by mean length of utterance in words (MLUw) and Total Number of Different Words (TNDW). MLU is the most widely used measure of grammatical productivity (Bedore, Peña, Gilllam, & Ho, Reference Bedore, Peña, Gillam and Ho2010), and MLUw is typically used when comparing languages with different morphological patterns (Goldstein, Bunta, Lange, Rodriguez, & Burrows, Reference Goldstein, Bunta, Lange, Rodriguez and Burrows2010). TNDW has been found to be positively associated with children's vocabulary by other lexical measures (Condouris, Meyer, & Tager-Flusberg, Reference Condouris, Meyer and Tager-Flusberg2003), and it is commonly used as an indicator of children's lexical development (Hewitt, Hammer, Yont, & Tomblin, Reference Hewitt, Hammer, Yont and Tomblin2005).
Since dual language learners are a heterogeneous linguistic group (Hoff & Core, Reference Hoff and Core2013) – some may have equal proficiency in both languages, others may be slightly more proficient in one of the two languages, and still others may show a significant dominance of one language over the other – we also conducted a preliminary cluster analysis to explore whether there were different profiles of dual language learners in terms of their language mixing rates, MLUw, and TNDW. Although cluster analysis has been deemed particularly useful for distinguishing various learner profiles, it has rarely been used in past research of dual language learners (Csizér & Dörnyei, Reference Csizér and Dörnyei2005). The objective of cluster analysis is to identify, within a given sample, certain subgroups – or clusters – of participants who share similar combinations of characteristics. The rationale for using this procedure in this study is the finding that language mixing patterns might be either positively or negatively related to proficiency in different children. That is, some learners with high utterance length and lexical diversity in English but low MLUw and TNDW in Spanish may show a low incidence of language mixing when conversing in English but high mixing rates when speaking Spanish (as found in Greene et al., Reference Greene, Peña and Bedore2012, Gutiérrez-Clellen et al., Reference Gutiérrez-Clellen, Simón-Cereijido and Leone2009, and Ribot & Hoff, Reference Ribot and Hoff2014). At the same time, the opposite profile (high MLUw and TNDW in Spanish but low MLUw and TNDW in English) might not result in high language switching rates when speaking English (Gutiérrez-Clellen et al., Reference Gutiérrez-Clellen, Simón-Cereijido and Leone2009; Ribot & Hoff, Reference Ribot and Hoff2014). Yet other children with balanced language skills across languages can show either high (Yow et al., Reference Yow, Patrycia, Flynn, Nicoladis and Montanari2016) or low language mixing rates (Ribot & Hoff, Reference Ribot and Hoff2014). Thus, we believe that cluster analysis could uncover different profiles of children with varied proficiency and language mixing patterns. While our approach is certainly innovative, we must point out that our analysis is purely exploratory – and our interpretation tentative – given the limited number of participants.
Method
Participants
The data for the present study were derived from a larger longitudinal study of bilingual development in Mexican-American dual language learners enrolled in Head Start programs in Southern California. We focused on the spontaneous speech samples of 26 randomly selected children (11 males, 15 females – the original sample included more females than males) collected in Spanish and English at age 3;6 (age range: 3;2–4;2), at the beginning of the program, and at age 4;5 (age range: 4;1–5;1), in the fall of their second year in the program. All children were born in the US but came from Mexican-origin families who had been living in Los Angeles County for an average of 12.7 years (SD = 8.7; Range = 6–23). All families came from low Socioeconomic Status (SES) backgrounds as determined by the children's eligibility to participate in the Head Start program, which is specifically designed to promote school readiness in children between birth and age five from low-income families (https://www.acf.hhs.gov/ohs). Indeed, as shown in Table 1, all mothers had limited education, with the majority of them having completed high school (76.9%) or only primary school (23.1%).
Participants’ gender, age at T1, mother's education, and parent-to-child, child-to-parent, and sibling language use
All children were typically developing and had no hearing, speech, language, cognitive, or neurological deficits based on parental reports and program screening tests. In order to determine the children's language history, exposure patterns, and language use, a detailed parent questionnaire was administered in Spanish. All parents reported a Mexican background/nationality and the Mexican Spanish variety as the language most spoken in the home. All children learned Spanish as their first language, since this was the language primarily used at home. At the same time, since all participants were born in the US, they had also had ‘some’ exposure to English from infancy through siblings, media, and the larger community, although formal and consistent exposure to this language had begun in preschool. In terms of language exposure and mixing patterns (see Table 1), the majority of parents indicated, on a scale from (1) to (4), that they spoke (1) only Spanish to their children (73.1%) or used (2) “mostly Spanish with a few English words” (26.9%), a result that suggests that exposure to code-switching through parental input was not widespread. However, language use was more mixed when children spoke to their parents (57.7%) and in sibling-to-sibling interactions (55.6%).
Classroom observations provided additional information about the language environment in the school. The Head Start program followed an English immersion model; for instance, for most children, formal and consistent exposure to English had begun in preschool. However, some teachers and classroom assistants were bilingual and addressed children in Spanish at times. An analysis of language use in the classroom using the Language Interaction Snapshot (LISn) (Atkins-Burnett, Sprachman, Lopez, Caspe, & Fallin, Reference Atkins-Burnett, Sprachman, Lopez, Caspe, Fallin, Howes, Downer and Pianta2011) – which captures the language use of teachers and adults during a period of five minutes divided into ten cycles of thirty-second segments each – indeed revealed that teachers spoke English for about 89% of the time and Spanish for about 11% of the time, with intra-sentential code-switching virtually non-existent in their speech. Thus, the participants in this study received input in, and had different degrees of proficiency in, Spanish and English, but Spanish tended to be the home and dominant language and English the school and weaker language. Code-switching was limited in the Spanish and English input children received from parents and teachers; however, it was more common in their interactions with siblings and in child-to-parent conversations.
Procedures
As in previous seminal studies of children's language mixing (Genesee et al., Reference Genesee, Boivin and Nicoladis1996; Genesee et al., Reference Genesee, Nicoladis and Paradis1995; Paradis & Nicoladis, Reference Paradis and Nicoladis2007), the data for this study were collected in naturalistic conversations in monolingual language contexts, where children were interacting with separate research assistants who, although bilingual, posed as monolingual English and Spanish speakers. While it is true that this methodology might discourage children from mixing languages, it was deemed necessary to examine children's ability to adhere to the interactional patterns of the school context (which was predominantly monolingual English) – and explore children's language behavior in a parallel monolingual Spanish setting – after formal and consistent exposure to English had begun in preschool and while children were being acculturated to the sociolinguistic characteristics of their environment.
Before collecting the speech samples, the research assistants became familiar with the children by spending time and interacting with them in their classroom. At the time of data collection, children were individually taken to a room during regular school hours and audio recorded for approximately 45 minutes in one language. The research assistants used a predetermined set of age-appropriate toys and books – including stuffed animals, dolls, cars, and books depicting common and culturally appropriate themes (going to school, attending a picnic, a birthday party, etc.) – to interact with the participants and elicit spontaneous speech. Since a monolingual-speaking situation was created, research assistants spoke only Spanish or English, depending on what language sample they were collecting. In situations when children spoke the non-target language, the research assistants pretended not to understand, and asked the children to repeat it in the target language. In total, four speech samples were collected from each child, a Spanish and English speech sample at age 3;6 and a Spanish and English speech sample at age 4;5.
Transcription and analyses
The samples were transcribed orthographically and analyzed by two additional research assistants using Systematic Analysis of Language Transcripts (SALT; Miller & Iglesias, Reference Miller and Iglesias2008). Each transcription was double-checked by a second transcriber and any disagreement was discussed by listening to the recordings several more times until consensus was reached. The consensus transcriptions were used for the analysis. The first 100 utterances of each sample (after skipping the first ten utterances) were then coded for intra-sentential and inter-sentential language mixing by the same two coders, trained on our criteria listed below, to ensure reliability and consistency of the coding. English samples were coded for Spanish language mixing and Spanish samples were coded for English language mixing. As shown in example (1) below, intra-sentential language mixing was defined as any mixing within the child's utterance, whereas inter-sentential language mixing was defined as the mixing of complete utterances in the non-target language as in Genesee et al. (Reference Genesee, Nicoladis and Paradis1995). Proper nouns in the non-target language and imitations were not counted as language switches.
(1) experimenter: What would you like to play?
child: I want the juguete (‘the toy’). [INTRA-SENTENTIAL MIX]
experimenter: The what?
child: ¿Por qué no me lo das? (‘Why won't you give it to me?’) [INTER-SENTENTIAL MIX]
child: ¡dámelo! (‘give it to me!’) [INTER-SENTENTIAL MIX]
experimenter: I don't speak Spanish.
child: Give it to me por favor! (‘please’) [INTRA-SENTENTIAL MIX]
experimenter: What is it that you want?
child: The car!
Inter-rater reliability of coding was performed by having a second transcriber independently re-code each sample for inter-sentential and intra-sentential language mixing. Disagreements were discussed by re-examining the transcriptions until consensus was reached.
The proportions of inter- and intra-sentential language mixing in each language context were then calculated out of the 100 utterances. It is important to mention that, while all children produced 100 utterances in both English and Spanish at age 4;5, seven children did not produce 100 utterances in English at age 3;6. This was not surprising since the children came from predominantly Spanish-speaking homes and had just begun preschool. In these cases, we calculated the proportion of language mixing out of the total number of utterances produced by the child in the English context. We eliminated one case from the analyses because the child (one female) produced 15 utterances, a database that was considered too limited to yield valid findings. The remaining children produced between 65 and 89 utterances in English at age 3;6, but their proportions of language mixing in this context were not statistically different from the proportions of language mixing in the same context by the children who produced 100 utterances (t(21) = 1.777, p = .071). Additionally, out of the remaining children, two other cases (two girls) were eliminated from the language mixing and cluster analyses because the z-score of the language mixing variables exceeded the 3.29 rule of thumb for outliers. This left us with data from 11 boys and 12 girls.
Mean length of utterance in words (MLUw) and Total Number of Different Words (TNDW) were automatically generated by SALT for the first 100 utterances (after skipping the first ten utterances). Recall that MLUw is typically used when measuring grammatical productivity in young bilinguals speaking languages with different morphological patterns (Goldstein et al., Reference Goldstein, Bunta, Lange, Rodriguez and Burrows2010), while TNDW is commonly used as an indicator of children's lexical development in both monolingual and bilingual children (Hewitt et al., Reference Hewitt, Hammer, Yont and Tomblin2005). MLUw in English was calculated from the pure English utterances and the English portion of language-mixed utterances. MLUw in Spanish was calculated from pure Spanish utterances and the Spanish portion of mixed utterances. Although some researchers completely eliminate language-mixed utterances when calculating MLUw (Yow et al., Reference Yow, Patrycia, Flynn, Nicoladis and Montanari2016), we considered this method too harsh and decided to give children credit for every word they produced in the target language. TNDW scores were calculated by counting English words only for English and Spanish words only for Spanish.
Statistical comparisons of patterns of language mixing, MLUw values, and TNDW scores in different languages and at different times were completed using the Statistical Package for the Social Sciences (SPSS Statistics Version 22). Paired-sample t-tests were used to examine whether language mixing patterns, MLUw, and TNDW scores changed across time and language. A univariate two-way analysis of variance (ANOVA) was conducted to assess whether time, language, and the interaction of time and language predicted language mixing patterns. In addition, correlations between proportions of language mixing and MLUw and TNDW values were also calculated at each time and in each language to assess whether such language alternation was related to proficiency in either language. We finally ran a cluster analysis to explore whether there were different profiles of children sharing similar patterns of language mixing, MLUw, and TNDW, using a series of one-way within-subjects ANOVAs to examine differences among clusters. The alpha level was adjusted for multiple comparisons throughout using the Bonferroni correction method (Holm, Reference Holm1979). Effect sizes were calculated using Cohen's d for t-tests and partial eta squared (η 2) for ANOVAs.
Results
Language mixing patterns at ages 3;6 and 4;5
Figure 1 displays the proportion of intra-sentential and inter-sentential language mixing in the English and Spanish context at ages 3;6 and 4;5. The results show that the proportion of language mixing was low in both languages and at both ages, in line with the results of previous studies (Gutiérrez-Clellen et al., Reference Gutiérrez-Clellen, Simón-Cereijido and Leone2009; Ribot & Hoff, Reference Ribot and Hoff2014). We first examined the intra-sentential and inter-sentential language mixing patterns of the participants at age 3;6 by conducting a set of four paired-sample t-tests. In order to protect against Type I error rate, we used the Bonferroni procedure, which resulted in the new critical alpha of .0125. The first t-test showed no significant difference between the proportion of intra-sentential (M = 0.05) versus inter-sentential (M = 0.06) language mixing of English into Spanish (d = .18). In contrast, there was a higher proportion of inter-sentential (M = 0.12) than intra-sentential (M = 0.03) language mixing of Spanish into English (t(22) = –2.86, p = .009, d = .60) (moderate effect size). In other words, while the proportion of language switches between and within sentences were similar when the children spoke Spanish, their use of Spanish in the English context predominantly consisted of complete Spanish utterances. However, cross-language comparisons revealed no significant differences between the proportion of inter-sentential and intra-sentential language mixing in the Spanish and English context at age 3;6.
Proportion of intra-sentential and inter-sentential language mixing in the Spanish and English context at ages 3;6 and 4;5.
A set of four paired-samples t-tests were again conducted to examine intra-sentential and inter-sentential language mixing patterns at age 4;5, keeping the critical alpha of .0125 to protect against Type I error. The first pair revealed a statistical significant difference between intra-sentential (M = 0.09) and inter-sentential (M = 0.05) language mixing of English into Spanish (M = 0.05) (t(22) = –3.46, p = .002, d = .72) (moderate effect size). However, there was no statistical significant difference between intra-sentential (M = 0.02) and inter-sentential (M = 0.00) mixing of Spanish into English at the same age (d = .34). Cross-language comparisons revealed that the participants produced intra-sentential language mixing at a higher incidence while speaking Spanish (M = 0.09) than while speaking English (M = 0.02) (t(22) = –4.31, p < .001, d = .89) (large effect size). Likewise, there was a higher proportion of inter-sentential language mixing in the Spanish context (M = 0.05) than in the English one (M = 0.00) (t(22) = 4.81, p < .001, d = 1) (large effect size).
Figure 2 displays the proportion of total language mixing in the English and Spanish context at ages 3;6 and 4;5. At 3;6, there was no statistically significant difference between the proportion of total language mixing in the Spanish context (M = 0.11) and the amount of total mixing in English (M = 0.16) (d = .28). At age 4;5, however, there was a higher proportion of language mixing when children spoke Spanish (M = 0.14) than when they spoke English (M = 0.03) (t(23) = 5.31, p < .001, d = 1) (large effect size). In other words, while the children switched language at about equal amounts in Spanish and English at age 3;6, a year later, they were inserting more English into their Spanish speech than Spanish into their English speech. Longitudinal analyses confirmed that the amount of English language mixing in the Spanish context did not change over the course of one year. However, there was a statistically significant difference in the proportion of Spanish language mixing into English between 3;6 (M = 0.16) and 4;5 (M = 0.03) (t(23) = 3.87, p = .001, d = .79) (large effect size). These results indicate that using Spanish while speaking English became virtually non-existent after a year of preschool.
Proportion of total language mixing (intra-sentential and inter-sentential combined for each language) in the Spanish and English context at ages 3;6 and 4;5.
MLUw and TNDW scores at ages 3;6 and 4;5
Figure 3 shows MLUw and TNDW scores at ages 3;6 and 4;5. Paired samples t-tests (keeping the critical alpha of .0125 to protect against Type I error) revealed statistically significant higher MLUw values in Spanish at age 4;5 (M = 2.93) than at age 3;6 (M = 2.40) (t(24) = –5.06, p < .001, d = 1) (large effect size). Similarly, MLUw in English was higher at age 4;5 (M = 3.09) than at age 3;6 (M = 2.09) (t(24) = –6.29, p < .001, d = 1) (large effect size). In addition, cross-language comparisons revealed a statistically significant difference between the children's initial English (M = 2.09) and Spanish (M = 2.40) MLUw at age 3;6 (t(24) = 3.15, p = .004, d = .63) (moderate effect size). However, at age 4;5, the participants’ MLUw in Spanish (M = 2.93) was not statistically different from their MLUw in English (M = 3.09) (d = .14). This means that, although the children were initially dominant in Spanish, after one year of preschool they had, on average, similar proficiency levels in both languages.
Mean Length of Utterance in words and Total Number of Different Words in English and Spanish at ages 3;6 and 4;5.
These results were confirmed by the TNDW analyses. For instance, at age 3;6, there was a statistically significant higher incidence of Spanish unique words (93.76) than English ones (68.32) (t(24) = 3.45, p = .002, d = .69) (moderate effect size). At age 4;5, however, the children produced comparable amounts of unique words in Spanish (M = 110.56) and in English (M = 109.00) (d = .07). Growth was evident in both languages. Indeed, there was a higher incidence of unique Spanish words at age 4;5 (M = 110.56) than age 3;6 (M = 93.76) (t(24) = –4.78, p < .001, d = .96) (large effect size). Similarly, the children produced more English unique words at age 4;5 (M = 109.00) than at age 3;6 (M = 68.32) (t(24) = –5.58, p < .001, d = 1) (large effect size). Overall, these results suggest that the children made gains in proficiency in both languages over the course of one year; however; growth was more evident in English than in Spanish, as found in other studies of young dual language learners growing up in the US (Hoff, Quinn, & Giguere, Reference Hoff, Quinn and Giguere2018; Ruiz-Felter, Cooperson, Bedore, & Peña, Reference Ruiz-Felter, Cooperson, Bedore and Peña2016; Scheffner-Hammer, Lawrence, & Miccio, Reference Scheffner-Hammer, Lawrence and Miccio2008).
Language mixing as related to age, language, and proficiency
A series of 2 (language) by 2 (age) ANOVAs was used to evaluate main effects and language by age interaction effects on inter-sentential, intra-sentential, and total language mixing. The alpha level was again adjusted using the Bonferroni correction (Holm, Reference Holm1979). The model indicated significant main effects for age for all three types of language mixing (inter-sentential: F(1,88) = 11.048, p = .001; intra-sentential: F(1,88) = 3.856, p = .045; total language mixing: F(1,88) = 5.705, p = .01). The effect sizes for inter-sentential, intra-sentential, and total language mixing were large to moderate (partial η 2 = .11, .042, .06, respectively) (Cohen, Reference Cohen1988). The model also indicated significant main effects for the age by language interaction (inter-sentential: F(1,88) = 8.765, p = .004; intra-sentential: F(1,88) = 7.446, p = .008; total language mixing: F(1,88) = 11.436, p = .001). The effect sizes for inter-sentential, intra-sentential, and total language mixing were also moderate to large (partial η 2 = .09, .08, .11, respectively). However, significant main effects for language were only found for intra-sentential mixing (F(1,88) = 32.926, p < .001), where the effect size was large (partial η 2 = .27). The marginal means for total language mixing in English were .19 (SE = .28) at 3;6 and .03 (SE = .28) at 4;5, a statistically significant mean difference of .16 (95% CI [.08, .24], p < .001). Similarly, the marginal means for language mixing at age 4;5 were .14 (SE = .28) in Spanish and .03 (SE = .28) in English, a statistically significant mean difference of .11 (95% CI [.04, .19], p = .005). These results suggest that language mixing patterns differed depending on age and language, with all types of mixing decreasing in English but not in Spanish as the children got older.
In order to investigate whether language mixing patterns in the Spanish and English context were linked to proficiency in Spanish and English as measured by MLUw and TNDW in each language, a series of correlation analyses was carried out separately for ages 3;6 and 4;5. Since the results did not differ between intra-sentential and inter-sentential switches, we calculated correlations using the proportions of total language mixing. Interestingly, the results showed few links between most proficiency measures and language mixing patterns and the only significant relationships included TNDW in English and language mixing in both contexts at age 3;6. In particular, we found that the proportion of total mixed utterances in the English context at age 3;6 was negatively correlated with TNDW in English (r = –0.65, p < .001, adj. α = .0125), suggesting that children who mixed more to Spanish while speaking English had smaller English vocabularies. Similarly, we found a moderate, positive relationship between language mixing patterns in Spanish and TNDW in English at the same age (r = 0.61, p = .04, adj. α = .0125), a finding that might indicate that more language mixing in the Spanish context was related to larger English vocabularies.
Exploratory cluster analysis
An exploratory cluster analysis was run on 23 cases, on English and Spanish variables corresponding to language mixing (total), MLUw, and TNDW at ages 3;6 and 4;5. Given that the aforementioned variables differed in scales, we computed their z-scores and used them for this analysis. A hierarchical cluster analysis using Ward's method produced three clusters between which most of the variables were significantly different in their mean. See Figure 4 for details.
Cluster analysis of the participants as a function of MLUw, TNDW, and language-mixing (LM) variables at 3;6 and 4;5.
The first cluster (N = 8) (LOW BILINGUAL SKILLS / VARIABLE MIXING) was characterized by dual language learners who were less than one standard deviation below the mean on all proficiency and language mixing variables with the exception of language mixing in the English context at age 3;6 and in the Spanish context at age 4;5. As shown in Figure 4, these were possibly children with average to below-average skills in both languages (as compared to the other participants), although Spanish was their dominant language at both ages. These were also children who switched to Spanish while speaking English at preschool entry; however, a year later, they produced more language mixing while interacting in Spanish than when interacting in English. The second cluster (N = 6) (HIGH BILINGUAL SKILLS / LOW MIXING) was characterized by participants who were up to a standard deviation above the mean on Spanish and English TNDW and MLUw at both ages but their proportion of language mixing in English at age 3;6 and Spanish at both ages tended to be more than half a standard deviation below the mean. It thus appears that these children had above-average proficiency in both languages as compared to the other participants, and tended to produce minimal language mixing, especially at age 4;5. Finally, Cluster 3 (N = 9) (LOW SPANISH SKILLS / HIGH MIXING IN SPANISH) was characterized by dual language learners with Spanish TNDW and MLUw values that were between a half and one standard deviation below the mean at both ages. Yet, their language mixing proportion were between a half and one standard deviation above the mean at both 3;6 and 4;5, especially in the Spanish context. It thus appears that these children were stronger in English and they tended to mix more when interacting in Spanish.
In order to examine whether the clusters differed significantly from one another, a series of one-way within-subjects ANOVAs was conducted. The results revealed a statistically significant difference between the clusters in Spanish lexical diversity at both ages 3;6 (F(2,20) = 12.101, p < .001, partial η 2 = .35) and 4;5 (F(2,20) = 8.78, p = .002, partial η 2 = .32), with large effect sizes (Cohen, Reference Cohen1988). Similarly, there was a statistically significant difference between the clusters in English lexical diversity at both ages 3;6 (F(2,20) = 10.07, p = .001, partial η 2 = .33) and 4;5 (F(2,20) = 4.57, p = .023, partial η 2 = .24). The strength of these differences, as indexed by η 2, was also large. Analyses also indicated that the clusters differed significantly from one another on both Spanish MLUw (F(2,20) = 10.81, p = .001, partial η 2 = .34) and English MLUw (F(2,20) = 19.02, p < .001, partial η 2 = .40) at age 3;6. This pattern remained constant at age 4;5 for both Spanish MLUw (F(2,20) = 15.55, p < .001, partial η 2 = .38) and English MLUw (F(2,20) = 3.93, p = .036, partial η 2 = .22). The magnitude of effect sizes was large in all analyses. Furthermore, the results revealed that the clusters differed significantly from one another in the amount of language mixing in the Spanish context at age 3;6 (F(2,20) = 5.23, p = .004, partial η 2 = .30), and there was also a marginal statistical difference between the clusters in the mixing of Spanish into English at the same age (F(2,20) = 3.35, p = .055, partial η 2 = .20), with large effect sizes. However, by age 4;5, the clusters did not significantly differ from each other in the proportion of language mixing in the Spanish (F(2,20) = 2.32, p n.s., partial η 2 = .16) or English context (F(2,20) = 1.20, p n.s., partial η 2 = .11).
To identify specific differences, Tukey post hoc tests were carried out on the data. These results revealed that Cluster 1 (LOW BILINGUAL SKILLS / VARIABLE MIXING) differed from Cluster 2 (HIGH BILINGUAL SKILLS / LOW MIXING) on all lexical diversity and utterance length variables (TNDW in Spanish at 3;6, p = .025; TNDW in Spanish at 4;5, p = .017; TNDW in English at 3;6, p = .001; TNDW in English at 4;5, p = .018; Spanish MLUw at 3;6, p = .034; Spanish MLUw at 4;5, p = .003; English MLUw at 3;6, p = .000; English MLUw at 4;5, p = .03), but not on language mixing. Similarly, Cluster 2 (HIGH BILINGUAL SKILLS / LOW MIXING) was statistically different from Cluster 3 (LOW SPANISH SKILLS / HIGH MIXING IN SPANISH) on all Spanish proficiency measures (TNDW in Spanish at 3;6, p = .000; TNDW in Spanish at 4;5, p = .001; Spanish MLUw at 3;6, p = .000; Spanish MLUw at 4;5, p = .000; English MLUw at 3;6, p = .000) but not on most English measures or language mixing. On the other hand, Cluster 1 (LOW BILINGUAL SKILLS / VARIABLE MIXING) differed from Cluster 3 (LOW SPANISH SKILLS / HIGH MIXING IN SPANISH) only on TNDW in English at 3;6 (p = .008) and on language mixing in the Spanish context at 3;6 (p = .003). While this analysis is purely exploratory given the limited number of participants, it supports the hypothesis that there were three subgroups of participants that differed in their degree of bilingual competence, with Cluster 2 participants being proficient in both languages, Cluster 1 children having limited skills in both languages (as compared to the other participants), and Cluster 3 learners displaying more limited competence in Spanish. Moreover, some participants (Cluster 3) knew more words in English at age 3;6 and hence mixed more to English while interacting in Spanish, as compared to others (Cluster 1).
Discussion
This study analyzed the language mixing patterns of 26 Spanish–English dual language learners at preschool entry, at age 3;6, and a year later at age 4;5. Since the participants were from predominantly Spanish-speaking homes, we were interested in observing their language mixing behavior in monolingual English and Spanish contexts after formal and consistent exposure to English began in preschool. We examined: (1) the incidence of intra-sentential, inter-sentential, and total language mixing over the course of one year; (2) whether these patterns were related to proficiency as measured by MLUw and TNDW; and (3) whether there were different subgroups of children, among the participants, with similar patterns of language mixing, MLUw, and TNDW.
Our exploration of language mixing patterns in both languages between 3;6 and 4;5 revealed interesting and somewhat unexpected results. First of all, as documented in previous studies (Gutiérrez-Clellen et al., Reference Gutiérrez-Clellen, Simón-Cereijido and Leone2009; Paradis & Nicoladis, Reference Paradis and Nicoladis2007), the frequency of language mixing was low in both languages and at both ages in the context of English and Spanish monolingual settings. Indeed, only between 11% and 16% of children's utterances at age 3;6 contained language switches; at age 4;5, language mixing remained around 14% in the Spanish context but decreased to 3% when children interacted with English-only speakers. In line with the predictions of the Control Process Model of code-switching (Green, Reference Green2018; Green & Wei, Reference Green and Wei2014), which posits that the environment – and the interactional context – dictate code-switching behavior, these results suggest that young preschoolers are good at choosing the language that is appropriate to the language context and at adhering to it for most of their conversational turns. The findings also support the hypothesis that bilingual children have increased communicative competence (Comeau et al., Reference Comeau, Genesee and Lapaquette2003; Reyes, Reference Reyes2004) because they are able to monitor patterns of language use in their environment and adjust their own accordingly, showing sensitivity to their interlocutors’ language use and preferences.
Interestingly, despite coming from primarily Spanish-speaking homes and being dominant in Spanish, the children showed similar proportions of language mixing while speaking Spanish and English at preschool entry. The intra- and inter-sentential language mixing analysis revealed that the children tended to insert complete Spanish utterances in their English, rather than single words, a type of language mixing that has been documented at early stages of bilingualism (Peynircioglu & Durungulo, Reference Peynircioglu, Durgunoglu, Hereida and Altarriba2002) or in less proficient bilinguals (Poplack, Reference Poplack1980). Other than that, the participants resorted to both languages at about equal amounts at age 3;6 and they also produced similar proportions of inter- and intra-utterance language mixing across languages. A year later, language mixing patterns had changed significantly in English but not in Spanish. In Spanish, children continued to mix languages to the same extent as they did at age 3;6, except that switches at the intra-utterance level became more common than those at the inter-utterance level, in line with studies that document more intra-sentential language mixing with increased bilingual proficiency (Poplack, Reference Poplack1980; Yow et al., Reference Yow, Patrycia, Flynn, Nicoladis and Montanari2016). In English, instead, language mixing decreased dramatically and using Spanish while speaking English became virtually non-existent after a year of preschool. Overall, these results suggest that language mixing patterns differed depending on language and age, with all types of mixing decreasing in English but not in Spanish as the children got older.
The analyses of the children's proficiency in Spanish and English as measured by MLUw and TNDW revealed that, as expected based on their language history, the children had stronger grammatical and lexical skills in Spanish at preschool entry. However, over the course of one year, the participants made gains in proficiency in both languages, with growth in English being more evident than growth in Spanish. This finding confirms the results of previous studies that document an accelerated growth in English after entering preschool but a deceleration in the development of Spanish among dual language learners growing up in the US (Hoff et al., Reference Hoff, Quinn and Giguere2018; Ruiz-Felter et al., Reference Ruiz-Felter, Cooperson, Bedore and Peña2016; Scheffner-Hammer et al., Reference Scheffner-Hammer, Lawrence and Miccio2008). Interestingly, the results showed no relation between most proficiency measures and language mixing patterns, with the exception of TNDW in English and language choice at age 3;6. In particular, at preschool entry, English lexical skills were negatively related to language mixing in the English context but they were positively related to language alternation in the Spanish context. These findings suggest that children with smaller English vocabularies mixed more to Spanish when speaking English, but they produced fewer switches to English when speaking Spanish. Yet no links were found between grammatical skills and language mixing patterns or between proficiency measures and language choice at age 4;5.
The exploratory cluster analysis revealed three subgroups of participants that differed in proficiency and language mixing patterns – a finding that was masked by the analyses that focused on means. Among the participants, there seemed to be about a third – dominant in Spanish but with average to below-average skills in both languages as compared to the other participants (Cluster 1, LOW BILINGUAL SKILLS / VARIABLE MIXING) – who switched to Spanish when interacting in their non-dominant language, English, at age 3;6, but who mixed more towards English at age 4;5 after a year of acculturation to the language of the school. On the other hand, another 25% of children (Cluster 2, HIGH BILINGUAL SKILLS / LOW MIXING) had above-average proficiency in both languages as compared to the other participants, and produced minimal amounts of language mixing at both ages, as the participants in Ribot and Hoff (Reference Ribot and Hoff2014) with similar expressive and receptive skills in English and Spanish. The rest of the participants (Cluster 3, LOW SPANISH SKILLS / HIGH MIXING IN SPANISH) were stronger in English and they tended to mix more when interacting in Spanish, their weaker but also the minority language, as the English-dominant participants in Ribot and Hoff (Reference Ribot and Hoff2014). While our analysis is purely descriptive – and our interpretation speculative – given the limited number of participants, these findings highlight the variability that characterizes young bilinguals’ language abilities and patterns of language use, and support the notion that proficiency and sociolinguistic factors might interact together to produce different dual language learners’ profiles.
It is possible that children's language mixing is constrained by proficiency – i.e., lexical gaps – only at early stages of bilingual development, when children have unbalanced language skills or have not reached a ‘threshold’ of bilingual competence (Paradis & Nicoladis, Reference Paradis and Nicoladis2007). For instance, the incidence of switches to Spanish in the English context was related to English lexical skills – and possibly due to vocabulary gaps – only at age 3;6, when children were, on average, more proficient in Spanish than in English. The cluster analysis further revealed that the three subgroups of participants differed in their use of language mixing only at preschool entry, with children with more limited language skills in one (Cluster 3, LOW SPANISH SKILLS / HIGH MIXING IN SPANISH) or both languages (Cluster 1, LOW BILINGUAL SKILLS / VARIABLE MIXING) producing more language mixing in their non-dominant language than those children with above-average and more balanced proficiency in both languages (Cluster 2, HIGH BILINGUAL SKILLS / LOW MIXING). These results seem to confirm the hypotheses that language mixing in young bilinguals is the result of asymmetries in proficiency levels (Greene et al., Reference Greene, Peña and Bedore2012; Ribot & Hoff, Reference Ribot and Hoff2014), and it is developmental and characteristic of the early phases of bilingual learning.
However, language mixing may cease to be proficiency-related with increased bilingual competence. Indeed, at age 4;5, when children's English and Spanish skills were comparable on average, language mixing patterns were no longer related to language ability and they did not differ for the three clusters of participants. In Spanish, children continued to mix languages to the same extent as they did at preschool entry; in fact, they switched more at the intra-utterance level as documented among bilinguals with increased bilingual competence (Poplack, Reference Poplack1980; Yow et al., Reference Yow, Patrycia, Flynn, Nicoladis and Montanari2016). In English, on the other hand, language mixing decreased dramatically for all participants, and using Spanish while speaking English became virtually non-existent. Even children who were still dominant in Spanish at age 4;5 (Cluster 1, LOW BILINGUAL SKILLS / VARIABLE MIXING) did not use Spanish while interacting in English. These findings indicate, for all children, an asymmetry in the direction of language mixing, with the Spanish context eliciting more switching to English than the English context eliciting mixing to Spanish. These results are consistent with other studies of Spanish–English dual language learners in the US that found that young bilinguals more frequently choose to speak English than Spanish (Oller & Eilers, Reference Oller and Eilers2002), and more frequently code-switch to English than to Spanish (Greene et al., Reference Greene, Peña and Bedore2012; Gutiérrez-Clellen et al., Reference Gutiérrez-Clellen, Simón-Cereijido and Leone2009; Ribot & Hoff, Reference Ribot and Hoff2014).
We speculate that, by this age, as the participants of Gutiérrez-Clellen et al. (Reference Gutiérrez-Clellen, Simón-Cereijido and Leone2009) and of Paradis and Nicoladis (Reference Paradis and Nicoladis2007), the children in this study had become aware of the sociolinguistic characteristics of their environment, the typical patterns of language use in their community, and the status of their languages. That is, they had learned that English was the majority language, the language with the highest status, and the preferred language in the community and in the school context where they were assessed. Thus, they found it appropriate to switch to English while speaking Spanish irrespective of their interlocutor's preference or language use. On the other hand, the children had developed awareness that Spanish was a minority language spoken only at home and by a segment of speakers in their community, a language “disfavored in typical American classrooms” and “within a sociopolitical context that does not support maintenance of the home language” (Gutiérrez-Clellen et al., Reference Gutiérrez-Clellen, Simón-Cereijido and Leone2009, p. 105). For this reason, they completely refrained from using it in the school context. These results are consistent with those of Garcia, Maez, and Gonzalez (Reference Garcia, Maez and Gonzalez1983), who also found that switching to Spanish in children's English discourse decreased as children became acculturated to the language of the school, whereas switching to English in Spanish discourse remained constant across grade levels. Taken together, and in line with the proposal of the Control Process Model of code-switching (Green, Reference Green2018; Green & Wei, Reference Green and Wei2014) that contextual factors dictate code-switching behavior, the results of this study indicate that, once children are sufficiently acculturated to the sociolinguistic characteristics of their environment, language choice may be more constrained by their awareness of the language prescribed by the majority culture than by their proficiency level in that language.
Conclusion, limitations, and future research
In conclusion, the results of this study underscore the complex interplay between proficiency-related variables and sociolinguistic factors when interpreting young bilinguals’ language choice. In the context of dual language learners in the US, language mixing may be initially related to limited language proficiency; but within a relatively short period of time, and with acculturation, children soon learn that using only English in a majority language context or switching to English while speaking the minority language are typical patterns of language use in their community. The results of this study also highlight the variability that characterizes dual language learners and their dual language use as they enter the school system. Some children may be dominant in Spanish, others in English; some children may have limited language skills in both languages, while others may have high levels of bilingual proficiency; some children may engage in language mixing quite freely, while others may refrain from mixing languages altogether. Taken together, these results suggest that grouping participants together and analyzing measures of average may capture some trends, but they may also obscure meaningful differences and fail to reveal different learners’ profiles.
Besides being of interest to language mixing scholars, the results of this study have practical implications. As dual language learners enter the school system, they are often at risk of being over- or under-identified for special education services (Kraemer & Fabiano-Smith, Reference Kraemer and Fabiano-Smith2017) because, in the English classroom, they might not be as intelligible as their monolingual peers and their practice of alternating between their two languages may be misinterpreted as a sign of confusion or of lack of proficiency or communicative competence (Hughes et al., Reference Hughes, Shaunessy, Brice, Ratliff and McHatton2006). The results of this study, however, inform educators, clinicians, and service providers that language mixing may be more common at early stages of bilingual proficiency, when children have not reached a ‘threshold’ of bilingual competence (Paradis & Nicoladis, Reference Paradis and Nicoladis2007). However, once children are sufficiently acculturated to the sociolinguistic characteristics of their environment, language choice may be more constrained by sociolinguistic variables – their awareness of the language prescribed by the majority culture – than by their proficiency level in that language. The findings of this study also point to the heterogeneity of dual language learners and the futility of explaining complex linguistic behaviors through simplistic, single-variable accounts.
While this study advances our understanding of language mixing in young dual language learners, it also has several limitations that should be considered when interpreting the results. First, the sample in this study was of moderate size, and language mixing and proficiency patterns were only assessed at two points in time, separated by one year. Clearly, we were only able to document developmental changes that occurred within this time frame. Future studies should examine language mixing in naturalistic conversations in larger samples and over a longer period of time, possibly documenting changes from age two, when home language should be even more dominant, to age three, when exposure to English in preschool begins, to the outset of formal schooling in kindergarten. It is particularly important that future studies be longitudinal because only this methodology can reveal whether early language mixing develops into the systematic and rule-governed skill of code-switching as children become increasingly proficient in their languages (Jisa, Reference Jisa2000).
Future investigators should also take into account that there may be different profiles of dual language learners in terms of their bilingual language development trajectory, and grouping participants together may mask important differences. While the cluster analysis conducted in this study was purely exploratory, future instigations should use this type of analysis with larger, heterogeneous samples to uncover different profiles of dual language learners sharing similar patterns of bilingual competence and use. Furthermore, since limited links were found between most proficiency measures and language mixing patterns, future studies should assess proficiency using different and perhaps more fine-grained measures, to examine whether other lexical and grammatical measures interact and predict language alternation at different points in development.
A final but important limitation of this study is that language mixing patterns were examined in monolingual language contexts, where children were interacting with research assistants who posed as monolingual speakers. Thus, the children were encouraged, if not required, to use only one language. While studies of language choice in young bilinguals have been traditionally carried out in these ‘monolingual’ contexts (Genesee et al., Reference Genesee, Boivin and Nicoladis1996; Genesee et al., Reference Genesee, Nicoladis and Paradis1995; Gutiérrez-Clellen et al., Reference Gutiérrez-Clellen, Simón-Cereijido and Leone2009; Paradis & Nicoladis, Reference Paradis and Nicoladis2007), it must be acknowledged that the goal of this research is to assess the extent to which children can adhere to a single language in discourse. Indeed, our own goal was to explore children's ability to adhere to the interactional patterns of the school context (which was predominantly monolingual English) – and of a parallel monolingual Spanish setting – after formal and consistent exposure to English had begun in preschool and while children were being acculturated to the sociolinguistic characteristics of their environment.
However, there is also a need to examine language mixing in bilingual contexts, where all parties speak the two languages and can freely switch from one to the other. Indeed, the studies that have interpreted language mixing as a sign of increased communicative competence have been carried out in contexts where children could freely engage in language mixing. For example, Comeau et al. (Reference Comeau, Genesee and Lapaquette2003) found that, when French–English bilingual toddlers were faced with unfamiliar interlocutors who code-mixed, they also mixed languages, adjusting their patterns of language use to those of their interlocutors. Similarly, Yow et al. (Reference Yow, Patrycia, Flynn, Nicoladis and Montanari2016), the only large-scale study of language mixing in a bilingual context to our knowledge, found that five-year-olds who came from a school in Singapore where Mandarin/English mixing was the norm, mixed languages at a significantly higher rate (32% of the time) than proficiency-matched children whose school discouraged code-switching (1.49%). Thus, monolingual contexts limit the opportunities for language mixing and cannot truly reveal the full range of factors that determine this linguistic behavior. Future studies should compare the language choices of the same children in both monolingual and bilingual contexts. Only such investigations can lead to a better understanding of the complex factors that affect language mixing in young bilinguals and reveal their true communicative competence while interacting with different speakers.
Acknowledgements
We acknowledge the support of grant #5SC3GM847583-3 to Kaveri Subrahmanyam, Marlene Zepeda, and Simona Montanari from NIGMS, and of grant #T34GM08228 to Wendy Ochoa from the MARC Undergraduate Student Training in Academic research program.
