1. Introduction
Bilingual children who are heritage language (HL) speakers acquire a minority language (HL) primarily at home, and a majority, societal language (SL) in the community/school, and come from families with a migration background. For sequential bilingual children, the HL is their first language (L1) and the SL is their second language (L2). Research on the development and maintenance of HLs has been growing in the past decade; nevertheless, most studies are cross-sectional and based on second-generation children, i.e., those born in the host country to first-generation/foreign-born parents (Montrul, Reference Montrul2016). Longitudinal studies with school-age children would offer a crucial link between early development and long-term outcomes because a shift to dominance in the SL for both proficiency and frequency of use is expected during the school years, along with the possibility of HL attrition or a plateau in growth (Carreira & Kagan, Reference Carreira and Kagan2011; Montrul, Reference Montrul2016, Reference Montrul2018, Reference Montrul2023). Longitudinal designs would reveal changes in children’s language environments, and in turn, the associations between language environment factors and individual differences in HL development (Paradis, Reference Paradis2023). Regarding immigration depth, research has found it to be associated with the quantity and quality of the HL home environment (Chondrogianni & Daskalaki, Reference Chondrogianni and Daskalaki2023; Daskalaki et al., Reference Daskalaki, Chondrogianni, Blom and Paradis2020). Because first-generation HL children were born in the home country and migrated to the host country at varying ages, their linguistic, community and educational experiences in the early years could be quite different from those of second-generation HL children. Therefore, it is possible that first-generation children, especially older arrivals, might have different profiles of dominant language shift and HL maintenance than second-generation children (Hamann et al., Reference Hamann, Chilla, Abed Ibrahim and Fekete2020; Montrul, Reference Montrul2023).
The recent resettlement of thousands of Syrian families in Canada (Immigration & Canada, Reference Immigration and Canada2023) has offered the opportunity to examine HL development prospectively over time in first-generation, Arabic-English bilingual children in the first five years post-resettlement. More specifically, the objectives of this longitudinal study were to determine: (1) if a shift in dominance from the Arabic-HL to the English-SL in terms of proficiency and language environment was taking place, as modulated by age of arrival, and (2) how language environment factors predicted individual variance in children’s Arabic-HL abilities after their initial years of residency in Canada.
2. Dominant language shift and development in heritage bilinguals
Dominant language shift refers to the relative strength of the HL and the SL in heritage bilinguals, and it is measured through comparisons of proficiency or frequency of use between the HL and SL over time. Regarding HL development, we use attrition to refer to decreases in skills overtime, growth to refer to increases in skills, and plateau to refer to skills that remain stable. The vast majority of bilingual children from migration backgrounds display a shift from the HL being their dominant language during the early years to the SL being the dominant language later on (Carreira & Kagan, Reference Carreira and Kagan2011; Montrul, Reference Montrul2016). Dominant language shift has been observed in both proficiency (Hamann et al., Reference Hamann, Chilla, Abed Ibrahim and Fekete2020; Pham & Kohnert, Reference Pham and Kohnert2014) and frequency of use (Carreira & Kagan, Reference Carreira and Kagan2011; Jia & Aaronson, Reference Jia and Aaronson2003). For simultaneous bilinguals, dominance shifts to the SL and slower growth in the HL can be observed before the age of 5;0, as shown in studies with Spanish-HL, English-SL bilinguals (Hammer et al., Reference Hammer, Scarpino, Davison, Neumann and Dickinson2011; Hoff & Ribot, Reference Hoff and Ribot2017). Studies with early sequential bilinguals in Germany have found that Turkish and Russian HLs can grow from ages 3;0–5;0 (schooling in German begins at 3;0), although growth rates and outcomes vary according to the HL, linguistic domain, and expressive versus receptive skills (Czapka et al., Reference Czapka, Topaj and Gagarina2021; Gagarina & Klassert, Reference Gagarina and Klassert2018). The age at which dominant language shift occurs for older sequential bilinguals varies according to linguistic domain, expressive versus receptive skills and also home language environment factors, but typically, this shift emerges during the first three years of elementary school for children with Arabic, Mandarin, Spanish or Vietnamese as HLs who began their schooling in the host country (Hamann et al., Reference Hamann, Chilla, Abed Ibrahim and Fekete2020; Jia & Aaronson, Reference Jia and Aaronson2003; Jia et al., Reference Jia, Chen, Kim, Chan and Jeung2014; Oppenheim et al., Reference Oppenheim, Griffin, Peña and Bedore2020; Pham & Kohnert, Reference Pham and Kohnert2014).
With respect to studies of dominance shift in first-generation bilinguals, findings are more limited and conflicting. A cross-sectional study of Arabic-HL refugee children aged 6;0-12;0 in Germany found that SL dominance was more common among second-generation children who had all their schooling in Germany (Hamann et al., Reference Hamann, Chilla, Abed Ibrahim and Fekete2020); however, the first-generation children had just 18 months of exposure to their German-L2, which might not have been sufficient time for a shift in dominance to take place. Jia and Aaronson (Reference Jia and Aaronson2003)’s longitudinal study of Mandarin-English bilinguals found evidence for a dominant language shift to the SL in language use preferences and abilities (via parent report) after three years of L2 exposure in the host country. By contrast, in a previous study with the same participant sample as the present study, Arabic-HL morphosyntactic skills remained superior to those in the English-L2 after nearly three years of L2 exposure (Paradis et al., Reference Paradis, Soto-Corominas, Daskalaki, Chen and Gottardo2021). The present study aimed to determine if Arabic dominance remained stable or shifted to English after four-and-a-half years of L2 exposure, in both relative proficiency and language use patterns. Finally, while it is often assumed that dominant language shift would go hand-in-hand with attrition, or plateauing of the non-dominant language (Montrul, Reference Montrul2016), Oppenheim et al. (Reference Oppenheim, Griffin, Peña and Bedore2020) did not find evidence for this among Spanish-HL bilinguals in English-only or transitional bilingual education programs (cf. Czapka et al., Reference Czapka, Topaj and Gagarina2021). It is possible that, for first-generation, school-age HL children, dominance shift and HL attrition might also be disentangled.
3. Age of arrival in heritage language development
In the case of first-generation bilinguals, AOA refers to “age of arrival”, typically coinciding with the age of (pre)school entry in the host country, and thus the onset of L2 acquisition. AOA also indexes the length of time a child has been acquiring the HL in a functionally monolingual environment prior to migration; thus, it can be considered a macro-variable indexing cognitive maturity, HL abilities, linguistic experiences, and family language use patterns at the onset of L2 acquisition. AOA has been found to be a predictor of variation in HL development, with younger AOA associated with earlier dominant language shift to the SL, more HL attrition and differential acquisition patterns in the HL (Albirini, Reference Albirini2018; Armon-Lotem et al., Reference Armon-Lotem, Rose and Altman2021; Czapka et al., Reference Czapka, Topaj and Gagarina2021; Daskalaki et al., Reference Daskalaki, Soto-Corominas, Barisé, Paradis, Chen and Gottardo2023; Gagarina & Klassert, Reference Gagarina and Klassert2018; Garibi & Boers, Reference Gharibi and Boers2019; Jia & Aaronson, Reference Jia and Aaronson2003; Meir & Janssen, Reference Meir and Janssen2021; Montrul, Reference Montrul2016; Paradis et al., Reference Paradis, Soto-Corominas, Chen and Gottardo2020; Paradis et al., Reference Paradis, Soto-Corominas, Daskalaki, Chen and Gottardo2021; Rodina et al., Reference Rodina, Kupisch, Meir, Mitrofanova, Urek and Westergaard2020; Soto-Corominas et al., Reference Soto-Corominas, Daskalaki, Paradis, Winters-Difani and Al Janaideh2022). In a study of Arabic-English bilinguals in the United States, Albirini (Reference Albirini2018) found that differential acquisition patterns in Arabic-HL morphosyntax were more prevalent for children with AOAs older than 3;0 than those with AOAs younger than 3;0, indicating that even a small delay in AOA can impact HL acquisition. Gharibi and Boers’s (Reference Gharibi and Boers2019) study of Persian–English bilingual children and adolescents in New Zealand found that AOA was associated with the strength of Persian-HL lexical skills while concurrent HL environment factors were not, signalling a robust role of AOA in HL development. Furthermore, G. Jia and Aaronson (Reference Jia and Aaronson2003) found that the dominance shift from Mandarin-HL to English-SL in language preference and abilities was modulated by AOA such that, those who arrived in the United States as adolescents did not exhibit the same shift as those who arrived as children. Finally, previous studies with the same participant sample as the present study found an older AOA advantage on morphosyntax production and comprehension tasks in Arabic at different time periods (Daskalaki et al., Reference Daskalaki, Soto-Corominas, Barisé, Paradis, Chen and Gottardo2023; Paradis et al., Reference Paradis, Soto-Corominas, Daskalaki, Chen and Gottardo2021). Accordingly, in this longitudinal study, the role of AOA was examined in HL development (growth, plateau, attrition) and dominant language shift.
4. Language environment factors in heritage language development
Language environment is a broad term encompassing both proximal input and interaction factors, as well as more distal factors that influence the proximal factors, and in turn, children’s language outcomes (Paradis, Reference Paradis2023). Proximal language environment factors include relative use of the SL and HL among family members and the frequency of engagement in language-rich activities in each language, outside of formal schooling in the SL. An important distal language environment factor for the HL in particular is access to literacy and/or education in this language. A key characteristic of language environment factors is that they are malleable, meaning they can change over time and be modified by intervention, in contrast to AOA. Previous research on dominant language shift suggests that the direction of change in children’s language environment would likely be from the HL to the SL (Carreira & Kagan, Reference Carreira and Kagan2011; Jia & Aaronson, Reference Jia and Aaronson2003)
Since HL is a minority language, the use of this language at home should be key to HL development. Indeed, more use of the HL at home is a positive predictor of HL abilities in studies with children aged 3;0 to 16;0 across a range of HL and SL contexts (Daskalaki et al., Reference Daskalaki, Chondrogianni, Blom and Paradis2020; Gagarina & Klassert, Reference Gagarina and Klassert2018; Jia et al., Reference Jia, Chen, Kim, Chan and Jeung2014; Jia & Paradis, Reference Jia and Paradis2020; Pham & Tipton, Reference Pham and Tipton2018; Rodina et al., Reference Rodina, Kupisch, Meir, Mitrofanova, Urek and Westergaard2020; Torregrossa et al., Reference Torregrossa, Caloi, Listani and Romano2023a). However, the language of sibling interaction could be particularly relevant as Czapka et al. (Reference Czapka, Topaj and Gagarina2021) found that HL input from siblings explained more variance in Russian-HL and Turkish-HL vocabulary than HL input from parents among 3 to 5-year-olds residing in Germany. In previous research with the participants in the present study, the use of more English overall and among siblings at home was not associated with HL abilities after two years of residency (Paradis et al., Reference Paradis, Soto-Corominas, Chen and Gottardo2020; Soto-Corominas et al., Reference Soto-Corominas, Daskalaki, Paradis, Winters-Difani and Al Janaideh2022), but it was a negative predictor of HL morphosyntax as children approached three years of residency (Paradis et al., Reference Paradis, Soto-Corominas, Daskalaki, Chen and Gottardo2021). The present study examines whether this negative association persists after a longer residency.
The richness of the HL environment should also be a key contributor to HL development given its minority status; however, fewer studies have examined richness, as compared to language use at home. Some studies with school-age children have found that more literacy and media engagement, extracurricular and cultural activities in the HL were predictive of stronger HL abilities in Mandarin and Vietnamese (Jia & Aaronson, Reference Jia and Aaronson2003; Jia & Paradis, Reference Jia and Paradis2015; Pham & Tipton, Reference Pham and Tipton2018). Previous studies with our refugee participants show mixed findings: we did not find associations between the richness of the HL environment as an aggregate variable (Paradis et al., Reference Paradis, Soto-Corominas, Chen and Gottardo2020; Paradis et al., Reference Paradis, Soto-Corominas, Daskalaki, Chen and Gottardo2021; Soto-Corominas et al., Reference Soto-Corominas, Daskalaki, Paradis, Winters-Difani and Al Janaideh2022); whereas, Daskalaki et al. (Reference Daskalaki, Soto-Corominas, Barisé, Paradis, Chen and Gottardo2023) used disaggregated richness variables and found that more frequent engagement in oral language activities in the Arabic-HL was predictive of children’s comprehension of Arabic syntax while other richness variables were not. Accordingly, disaggregated richness variables were examined in the present study. Finally, even though language environment factors are malleable and influence HL development, few studies have examined their changes longitudinally, e.g., Jia & Aaronson, Reference Jia and Aaronson2003, Oppenheim et al., Reference Oppenheim, Griffin, Peña and Bedore2020– with only Jia and Aaronson (Reference Jia and Aaronson2003) studying first-generation HL bilinguals. For Mandarin-HL children, Jia and Aaronson (Reference Jia and Aaronson2003) found more preference for the English-SL, increased richness in the English environment, and more use of English at home over time, as modulated by AOA; however, their small sample size and wide age range limit generalizations from this study, indicating more research is needed.
Possessing literacy skills in the HL and access to education in the HL, either as part of school programming or as complementary classes, has been found to increase HL lexical and morphosyntactic skills specifically, beyond the contribution to overall HL proficiency across a range of HL and SL contexts (Armstrong & Montrul, Reference Armstrong and Montrul2024; Bayram et al., Reference Bayram, Rothman, Iverson, Kupisch, Miller, Puig-Mayenco and Westergaard2017; Jia & Paradis, Reference Jia and Paradis2015; Kupisch & Rothman, Reference Kupisch and Rothman2018; Rodina et al., Reference Rodina, Kupisch, Meir, Mitrofanova, Urek and Westergaard2020; Torregrossa et al., Reference Torregrossa, Flores and Rinke2023b). The effect of education in the HL is expected to be multifold, encompassing increased quantity and quality of proximal input, as well as distal factors, such as supporting positive linguistic and cultural identity and attitudes (Montrul, Reference Montrul2023; Paradis, Reference Paradis2023). In particular, literate language is comprised of more advanced vocabulary and complex morphosyntactic structures and thus provides linguistic experiences in the HL beyond everyday conversation. Experiencing some education in the HL should be an advantage for first-generation, school-age migrant children over their second-generation peers. However, interrupted education is common among child refugees due to school closures during conflicts, migration journeys, and time in transitional countries and refugee camps where educational opportunities are not available (Fazel & Stein, Reference Fazel and Stein2002; Sirin & Rogers-Sirin, Reference Sirin and Rogers-Sirin2015). Some participants in our sample had interrupted schooling in Arabic before arrival in Canada (Paradis et al., Reference Paradis, Soto-Corominas, Vitroulis, Al Janaideh, Chen, Gottardo, Jenkins and Georgiades2022), but the association between the amount of schooling in Arabic and their Arabic-HL outcomes has not yet been investigated.
5. The present study
This study examined the Arabic-HL development of first-generation refugee children across 3 time periods (T1, T2, T3) during their first four-and-a-half years of residency in Canada. Arabic proficiency and frequency of use were compared to participants’ English-SL to address questions of dominance shift because dominance is a relative construct. Receptive vocabulary and sentence repetition tasks were administered at each time period in both languages, and parent questionnaires yielded the following age and language environment variables: AOA, language use in parent and sibling interactions, the richness of the language environment in both languages and the amount of schooling in Arabic. We included both a receptive (vocabulary) and an expressive (sentence repetition) task in this study for a more comprehensive view of language proficiency.
Arabic-speaking children grow up in a diglossic environment, where spoken varieties are used for informal contexts while Modern Standard Arabic (MSA) is the basis of literacy and used in formal contexts (Albirini, Reference Albirini2015). In our study, both tasks were administered in the spoken Syrian variety rather than MSA because it is the variety children use at home. Furthermore, using MSA would have disadvantaged the children who had little or no experience with schooling in Arabic, since schooling is the foundation for developing MSA skills (Albirini, Reference Albirini2015).
Our research questions were as follows:
-
(1) Is there is a shift in dominant language proficiency from Arabic to English from T1 to T3? Is there evidence of attrition or plateau in Arabic development? Does this differ as a function of younger versus older AOA?
-
(2) Are there shifts from Arabic to English as the dominant language of use at home or in language-rich activities from T1 to T3? Does this differ as a function of younger versus older AOA?
-
(3) What are the associations between language use at home, engagement in Arabic-rich activities, length of Arabic schooling pre-migration and participants’ Arabic outcomes, concurrently and longitudinally, at T3? This final question connects the findings from questions (1) and (2) to understand predictors of heritage language maintenance in first-generation bilinguals.
6. Methods
This longitudinal study included three data collection points: T1: December 2017–August 2018; T2: March–August 2019, and T3: October 2020–June 2021. The longer interval between T2 and T3 was due to the onset of the COVID-19 pandemic in March 2020, which necessitated a pivot to remote testing.
6.1. Participants
Participants and their families migrated to Canada as part of a government resettlement program for UNHCR-designated refugees from the conflict in Syria and resided in three cities: Edmonton, Waterloo and Toronto. All participants (T1: N = 133; mean age = 9;5; 67 females) spoke Syrian Arabic and no additional, non-English language; none of the participants had exposure to English prior to resettlement in Canada. Families moving away, families being untraceable, and families declining to participate contributed to the attrition rates at T2 and T3. Participant characteristics are presented in Table 1, Results, for each time interval.
6.2. Materials
6.2.1. Sentence Repetition Tasks (SRT; Soto-Corominas et al., Reference Soto-Corominas, Daskalaki, Paradis, Winters-Difani and Al Janaideh2022)
SRTs ask participants to listen to a sentence and to repeat it right away as accurately as possible. Performance on these tasks implicates lexical, morphosyntactic and speech production abilities, as well as verbal short-term memory (Polišenská et al., Reference Polišenská, Chiat and Roy2015; Pratt et al., Reference Pratt, Peña and Bedore2021). As such, sentence repetition tasks can index global proficiency, although they can be used to index morphosyntax in particular when analysed by the syntactic structure of items (Soto-Corominas et al., Reference Soto-Corominas, Daskalaki, Paradis, Winters-Difani and Al Janaideh2022). We employed verbatim scoring for the sentence repetition task for this study to index global proficiency in each language with this task.
6.2.2. English SRT
Stimuli were adapted for Canadian English from the English LITMUS SRT (Marinis & Armon-Lotem, Reference Marinis, Armon-Lotem, Armon-Lotem, de Jong and Meir2015). The items included a variety of syntactic structures: declarative sentences with modals and/or auxiliaries (k = 6), short and long passives (k = 6), wh-object questions (k = 6), sentences with coordinated clauses (k = 3), conditional/temporal subordinate clauses (k = 4), and object relative clauses (k = 6). The English SRT stimuli were recorded by a native speaker of Canadian English. The task had 32 sentences in total, with a mean of 10.03 morphemes per sentence. Each response was scored as correct if it was identical to the stimuli or incorrect otherwise. (For details on stimuli, see Soto-Corominas et al., Reference Soto-Corominas, Daskalaki, Paradis, Winters-Difani and Al Janaideh2022).
6.2.3. Arabic SRT
The Syrian Arabic version of the SRT was developed using the English SRT as a model but with some alterations in structures where English and Arabic are not compatible, e.g., long passives. The development process for the Arabic SRT was as follows: Stimuli were first translated into the spoken variety by a native speaker of the closely related variety of Jordanian Arabic. Subsequently, two native speakers of Syrian Arabic acted as consultants and changes were made to some items for the Syrian variety specifically; these changes were mainly lexical or phonological. The Arabic SRT included the following structures: declarative sentences with modals and/or particles (k = 6), short passives (k = 3), topicalizations (k = 3), wh-object questions (k = 6), sentences with coordinated clauses (k = 3), conditional/temporal subordinate clauses (k = 4), and object relative clauses (k = 6). The Syrian Arabic SRT stimuli were recorded by a native speaker of the Syrian variety of Arabic. The task had 32 sentences in total, with a mean of 10.06 morphemes per sentence. Each response was scored as correct if it was identical to the stimuli or incorrect otherwise. (For details on stimuli, see Soto-Corominas et al., Reference Soto-Corominas, Daskalaki, Paradis, Winters-Difani and Al Janaideh2022).
6.2.4. Arabic Language Assessment Battery - vocabulary subtest (ALAB; Assadi et al., Reference Assadi, Shany, Ibrahim, Khateb and Ben Simone2015)
Participants were asked to point at one picture out of an array of four that matched the word provided by the research assistant. All lexical items were based on the Levantine variety of spoken Arabic, which includes the varieties spoken in Syria. Some alternate stimuli were included for some words, based on input from our Syrian consultants. This test has 73 items and was designed for use with elementary school children. This test does not include standard scores, but comparison to monolingual norms was not a goal of this study.
6.2.5. Peabody picture vocabulary ttest- 4 (PPVT; Dunn & Dunn, Reference Dunn and Dunn2007)
The procedures for this test are the same as those for the ALAB. However, the PPVT was designed to be a lifespan test and has a total of 228 items. The difference in the total number of items/scales for the ALAB and the PPVT was taken into account in our analyses. While the PPVT does include standard scores, it was not a goal of this study to reference participant’s performance with monolinguals.
6.2.6. Alberta language environment questionnaire-4 (ALEQ-4; Paradis et al., Reference Paradis, Soto-Corominas, Chen and Gottardo2020)
We collected information on participants’ and families’ demographic characteristics and linguistic experiences via the ALEQ-4. AOA was determined by the age of school entry in Canada because families did not use English in the home before then, and the gap between arrival time and school entry was only around 2–3 months. Parents described the relative Arabic/English use in the home by using a 1–5 scale where higher numbers indicated more English use (1 = Mainly or only Arabic [English: 0%–20%, Arabic: 80%–100%], 2 = Usually Arabic/English sometimes [English: 30%, Arabic: 70%], 3 = Arabic and English [English: 50%, Arabic: 50%], 4 = Usually English/Arabic sometimes [English: 70%, Arabic: 30%], 5 = Mainly or only English [English: 80%–100%, Arabic: 0%–20%]). We collected this information in terms of output given to, and received from, the participants and their mother, father, older siblings, and younger siblings. Because families in our sample tended to be large, we decided to group older and younger siblings together to streamline the interview process.
We estimated the frequency of Arabic- and English-rich oral language and literacy activities by asking parents to indicate, for each language separately, how many hours per week participants spent (1) reading and writing (e.g., books, websites, messaging, homework), (2) speaking and listening (e.g., TV, streaming, music, video chatting), (3) using that language with friends, and (4) taking part in extra-curricular activities in that language (e.g., sports, clubs, cultural events, religious services). Parents indicated the frequency of these activities using an ordinal 1–5 scale: (1 = 0–1 h, 2 = 1–5 h, 3 = 5–10 h, 4 = 10–20 h, 5 = 20+ h).
6.3. Procedures
6.3.1. Data collection
The SRT and vocabulary task were delivered as part of a larger battery of linguistic and non-linguistic tests. Participants were tested in one or two sessions per language (for about 90 minutes total). The English and Arabic testing sessions occurred on different days, with language order counterbalanced across participants. The research assistants who carried out the testing sessions were native speakers of the given language.
Participants were tested individually at home or in a quiet room at their school for T1 and T2. Data collection at T3 was moved online due to the ongoing COVID-19 pandemic. The online data collection happened synchronously through the Zoom videoconference platform. Participants wore noise-cancelling headphones to carry out the entire testing session and the research assistant in charge of the testing session shared their computer screen, where the visual stimuli for the tests were displayed. For the SRT, some minor data loss occurred during online delivery due to technical issues like Wi-Fi interruptions. This affected some individual items on both the English and Arabic tasks: 2.4% loss of English items and 3.6% loss of Arabic items. Since the unscorable items were small in number and spread out among participants and items on the task, the individual participants in our study at T3 had zero to minimal data loss each. Data loss did not impact the vocabulary tests, which were not recorded or time-dependent, and were instead scored during the session by the research assistant.
The ALEQ-4 was administered as an interview to parents by an Arabic-speaking research assistant. Parents were interviewed in person or over the phone at T1 and T2, while at T3, parents were interviewed over the phone only.
6.3.2. Data analysis
All analyses were carried out in R (version 4.3.0; R Core Team, 2023). Given the large number of analyses required to address our research questions, for the sake of brevity, structural details of the models and summary model output information are presented in the Results section. Correlation matrices of the predictor and outcome variables for Arabic and English, full model outputs and results of post-hoc tests are available in the Supplementary Materials.
7. Results
7.1. Participant characteristics
Participant characteristics are displayed in Table 1. Summary data about language use at home and language richness factors are not presented in Table 1, as they are the focus of our analyses in the section, Dominance Shift in Language Use at Home and Language-Rich Activities. At T1, participants had a mean chronological age of 9;4, a mean of 20 months of exposure to English, and a mean English AOA of 7;7. Regarding maternal education, the mean of 9.67 years indicates that most mothers had not completed secondary school, signalling that the majority of families have low socioeconomic status (Paradis et al., Reference Paradis, Soto-Corominas, Vitroulis, Al Janaideh, Chen, Gottardo, Jenkins and Georgiades2022). Finally, even though the participants had a mean age of 9 years, their average length of education in Arabic was just 14.12 months, although with wide variation (SD = 14.85). As mentioned earlier, this indicates that many participants experienced interruptions in their education prior to resettlement.
For subsequent analyses, AOA was treated as a categorical variable, divided according to the median in our sample. Those who were classified as having a Young AOA had an AOA below the median for the group, and vice versa for the Old AOA group. The mean AOA for the Young group was 5;11 years (SD = 1;0), and the mean AOA for the Old group was 9;4 (SD = 1;3). We chose to operationalize AOA as a categorical variable for this study because of conceptual and statistical reasons. First, AOA indexes multiple cognitive, language experience and use, and HL proficiency factors. Such factors coalesce around broader categories of younger versus older arrivals, and we reasoned that meaningful differences in language ability and use are better captured by these categories than incrementally with increases in AOA as calculated in months. Second, treating AOA as a categorical variable permitted fitting models with random slopes for the SRT and vocabulary data, whereas models with AOA as a continuous variable could only be fitted with random intercepts. While models with AOA as a continuous variable produced similar overall results for our data, models with random slopes are better fitting and more appropriate for longitudinal data (Barr et al., Reference Barr, Levy, Scheepers and Tily2013). The relationships between AOA as a continuous variable and all other variables in this study can be examined in the correlation matrices, Figures S1 and S2, in Supplementary Materials.
Sample characteristics at the three time points
a Even though these variables remain stable over time, they are specified for the three time points due to differences in participant sample size causing minor changes in the numerical values.
7.2. Dominance shift in language proficiency and HL developmental trajectories
Our first research question asked about whether dominant language shift and HL attrition have taken place between T1 and T3, and whether either was modulated by AOA. Figure 1 shows the proportion of verbatim responses on the SRT for the Young and Old AOA groups. In order to address this question for the SRT data, we fit a binomial GLMER using the package lme4 (version 1.1-33; Bates et al., Reference Bates, Maechler, Bolker and Walker2015). Instead of modelling the total score, our analysis calculates the probability of being correct on each individual item. Doing so confers two main advantages: First, we are able to account for the portion of variance that is explained by individual items. Second, we avoid penalizing participants for the rare instances of experimental error, where an item was not played to the participant by mistake. As such, in the logistic regression model, the outcome variable was whether each item in the SRT had been repeated verbatim (1) or not (0). The fixed effects included a triple interaction between three factors: Time (T1, T2, T3), Language (Arabic, English), and AOA Group (Young, Old). We fit two random intercepts: one for Item and one for Participant, which was nested within Family because some sibling pairs were in the sample. In addition, we fit by-participant random slopes for Time and Language. (An interaction between Time and Language could not be fit as a random slope.) Post-hoc contrasts were analysed using the package emmeans (Lenth, Reference Lenth2022). Full model results and post-hoc contrasts are in Supplementary Appendix SA (S indicates these can be found in the Supplementary Materials).
Mean proportion of correct verbatim SRT repetitions for each AOA group (Young and Old) over Time (T1, T2, T3) in both Arabic and English. Error bars indicate standard errors.
The logistic regression model included a significant triple interaction between Time, Language, and AOA Group. The findings are as follows: At T1 and T2, participants were more accurate in Arabic than in English, regardless of the AOA group (all four contrasts: p < 0.001). However, at T3, the Young AOA group was more accurate in English than in Arabic (p = 0.010) and there were no differences by language for the Old AOA group (p = 0.899). Across all three time periods, the Old AOA group outperformed the Young AOA group in Arabic (all contrasts, p < 0.002), but that was not the case for English after T1 where group differences did not emerge (T1: p = 0.087 [trend], T2: p = 0.825, T3: p = 0.425). Finally, and as suggested by Figure 1, both Old and Young AOA groups made significant gains in English between T1 and T2, and between T2 and T3 (all contrasts, p < 0.001). On the other hand, for Arabic, both groups made gains between T1 and T2 (Old AOA: p < 0.001; Young AOA: p = 0.006), but for both groups, performance was significantly worse at T3 than at T2 (both contrasts p < 0.001).
Results for the Arabic and English vocabulary tests over time are in Figure 2. Since the tests had different scales (see Materials), plots are separated by language. Note that we could not provide a proportion score of correct responses out of the total number of items because of the design of the vocabulary tests. While the Arabic test is meant for elementary school children, the English test is a lifespan test, consequently, denominators for proportion scores would be very different so proportion scores would not be directly comparable. Since the two tasks could not be compared to each other, our analyses for the vocabulary task were aimed at finding the presence of plateauing or attrition, but not dominant language shift. We ran two separate logistic regressions, one for each language, where the accuracy for each item was predicted by Time (T1, T2, T3) and AOA Group (Young, Old), and the interaction between the two. The random-effects structure in these two vocabulary models included a random intercept by Item and one by Participant nested within Family, with a by-participant random slope for Time. (Full model results and post-hoc contrasts are in Supplementary Appendix SB).
Mean total scores for Arabic and English vocabulary tests over Time (T1, T2, T3) for each AOA group (Young and Old). Error bars indicate standard errors.
In the Arabic model, the interaction was not significant; however, AOA Group and Time were both significant predictors in an additive model without the interaction. Specifically, the Old AOA group outperformed the Young AOA group significantly (p < 0.001) overall. In terms of Time, there was a significant increase in performance between T1 and T2 (p < 0.001), but not between T2 and T3 (p = 0.880), according to post-hoc contrasts. In the English model, the interaction between AOA Group and Time was significant. According to post-hoc contrasts, both AOA groups made significant improvements between T1 and T2, and between T2 and T3 (all contrasts, p < 0.001). However, at T1, the Old AOA group outperformed the Young AOA group for English vocabulary (p = 0.021); this was not true at T2 (p = 0.928) or at T3 (p = 0.968).
7.3. Dominance shift in language use at home and language-rich activities
Our second research question asked whether changes in participants’ language environments over time showed a shift from Arabic to English, and if this differed according to AOA Group. To address this question, we fit cumulative link mixed models (CLMM) with the package ordinal (version 2022.11-16; Christensen, Reference Christensen2018). Because language use among family members and frequency of language-rich activities were measured using rating scales, this model was preferred for the correct treatment of ordinal-scale observations (i.e., ordered categorical data). All ordinal models presented below included one random intercept for Participant. For all models, backwards selection was followed by pruning the non-significant interaction terms. Where appropriate, post-hoc contrasts were conducted to probe differences across the three time periods.
7.3.1. Language use at home
Language use between participants and their parents, and participants and their siblings were analyzed separately. As a reminder, data are from relative language use scales where lower numbers mean more Arabic and higher numbers mean more English (1 = Mainly or only Arabic, 2 = Usually Arabic/English sometimes, 3 = Arabic and English, 4 = Usually English/Arabic sometimes, 5 = Mainly or only English). The initial models included the predictors Time, AOA Group, and Interaction Type (i.e., input to child or output from child), as well as the triple interaction between the three predictors. Model outputs and post-hoc results are in Supplementary Appendices SC and SD. Note that mothers’ and fathers’ scales were combined because there were no differences between them. As such, an average of the two ordinal scales was calculated in order to create the composite “Child input from parents” and “Child output to parents”. Similarly, we initially obtained measures of language use between participants and younger and older siblings separately; however, as no differences were found by sibling age, an average of two scales formed the composite variables “Child input from siblings” and “Child output to siblings”.
7.3.2. Language use with parents
Figure 3 (top panel) shows language use between parents and participants over time, which was mainly in Arabic given scores are between 1 and 2. All three predictors and the interaction between Time and AOA Group were significant (see Supplementary Appendix SC). The interpretation of this model is as follows: Participants, regardless of AOA Group, used significantly more English in their output to parents than they received in input from parents (p < 0.001). The Young AOA group, but not the Old AOA group, showed a significant overall increase in English use between T1 and T3 (p < 0.001). Both groups used English with parents to a similar extent at T1 (p = 0.593) and T3 (p = 0.154); whereas the Young AOA group used significantly more English than the Old AOA group at T2 (p = 0.034).
Mean language use between participants and their parents (top panel) and siblings (bottom panel), between 1 (only Arabic) and 5 (only English) according to AOA Group. Bars show standard errors.
7.3.3. Language use with siblings
Figure 3 (bottom panel) shows language use between participants and siblings over time, showing English trending towards the dominant language of interaction by T3 where scores were between 3.5 and 4.5. In the model, Time was the only significant predictor (Supplementary Appendix SD). Therefore, regardless of AOA group or interaction type, participants increased the use of English between T1 and T2, and between T2 and T3 significantly (p = 0.001 and p < 0.001, respectively) with the increase between T2 and T3 being steeper, judging by the slopes in Figure 3 and by the estimates of the pairwise contrasts.
7.3.4. Frequency of language-rich activities in Arabic and English
A series of four ordinal models examined frequency in an average week of reading/writing activities, listening/speaking activities, interactions with friends, and extracurricular activities in each language (1 = 0–1 h [never/almost never]; 2 = 1–5 h [a little]; 3 = 5–10 h [regularly]; 4 = 10–20 h [often]; 5 = 20+ h [very often]). Differently from the ordinal scales describing language use between participants and parents/siblings, these ordinal scales are not relative. That is, unlike language use, it is not the case that higher frequencies of activities in one language yield lower frequencies in the other language. For this reason, visualizations show the frequency ratings for Arabic and English separately (see Figure 4). In the ordinal models, the predictors were AOA Group, Time, and Language (Arabic, English). The initial model included the three predictors as well as a triple interaction between them. Full model outputs and post-hoc results are in Supplementary Appendices SE to SH.
Frequency of language-rich activities in each language according to AOA group and activity. The Y-axis shows a scale of frequency between 1 (0–1 h) and 5 (20+ h) in an average week. Bars are standard errors.
7.3.5. Reading/writing activities
The frequency of reading/writing activities for each language is displayed in the top panel of Figure 4. The optimal model included AOA Group, Time, Language, and the interaction between Time and Language (Supplementary Appendix SE). Specifically, the Young AOA group engaged less in reading/writing activities than the Old AOA group overall (p = .027). Both groups engaged more in reading/writing activities in English than in Arabic across time (all three contrasts p < 0.001), with the largest difference occurring at T3. The two languages also showed differences in terms of changes over time. While for Arabic, frequencies remained relatively low and stable over time (all three contrasts p > 0.05), for English, there was a significant increase between T2 and T3 (p = 0.019).
7.3.6. Speaking/listening activities
The frequency of speaking/listening activities for each language is displayed in Figure 4 (second panel). The optimal model included Time, Language, and the interaction between Time and Language as significant effects (Supplementary Appendix SF). No differences emerged between AOA groups. For Time, there was a significant decrease in the frequency of speaking/listening activities in Arabic between T1 and T2 (p = 0.002), but not between T2 and T3 (p = 0.935). In English, on the other hand, there was a significant and steady increase in the frequency of these activities across time, T1 and T2 (p = 0.018) and T2 and T3 (p < 0.001), with the latter change being much steeper, judging by both Figure 4 and the estimates of the post-hoc contrasts. For Language, the frequency of speaking/listening activities was significantly higher in Arabic than in English (p = 0.029) at T1. Conversely, at both T2 and T3, the frequency of speaking/listening activities was significantly higher in English than in Arabic (both contrasts p < 0.001), with the gap being especially large at T3.
7.3.7. Language use with friends
The frequency of language use with friends for each language is displayed in Figure 4 (third panel). The optimal model included Time, Language, and the interaction between Time and Language, as significant effects, with no differences between AOA Groups (Supplementary Appendix SG). For Time, in Arabic, the frequency of language use with friends was stable between T1 and T2 (p = 0.974), but there was a significant decrease between T2 and T3 (p < 0.001). In English, on the other hand, there was a significant increase in the frequency of English use with friends between T1 and T2 (p < 0.001), stabilizing between T2 and T3 (p = 0.727). Similar to the model for speaking/listening activities, the frequency of language use with friends was significantly higher in Arabic than in English (p = 0.005) at T1. However, at both T2 and T3, participants interacted with friends more frequently in English than in Arabic (both contrasts p < 0.001), with the gap being especially large at T3.
7.3.8. Extracurricular activities
The frequency of extracurricular activities for each language is displayed in Figure 4 (bottom panel), and it is evident that participants attended extracurricular activities infrequently, as scores fell between 1 and 2. The optimal model included Time, Language, and the interaction between Time and Language, as significant effects, with no differences between AOA Groups (Supplementary Appendix SH). For Time, in Arabic, the frequency of extracurricular activities did not change between T1 and T3 (all contrasts p > 0.05). In English, on the other hand, there were no changes between T1 and T2 (p = 0.122), but there was a significant decrease between T2 and T3 (p < 0.001). Second, the frequency of extracurricular activities was significantly higher in Arabic than in English at T1 (p = 0.005) and T3 (p < 0.001); however, no between-language differences emerged at T2 (p0.332).
7.4. Associations between language environment factors and Arabic outcomes at T3
Our third research question concerned how HL outcomes at T3 were associated with HL environment factors that change over time (Language use with siblings, Language use with parents, Frequency of activities in the HL [Reading/writing in Arabic, Speaking/listening in Arabic, Interactions with friends in Arabic, Extracurriculars in Arabic]), together with the stable factor of Length of schooling in Arabic pre-migration. Note that, in choosing to include the Length of schooling in Arabic prior to migration we could not include the AOA Group as a predictor, as both variables align with a similar construct (i.e., most of the participants with some schoolings were in the Old AOA group; the correlation between AOA and schooling in Arabic is r = 0.654). Therefore, we ran two logistic mixed-effects regression models, one for the Arabic SRT and one for Arabic vocabulary, where item accuracy for each test was predicted by the aforementioned variables. In addition to concurrent models, we conducted two longitudinal analyses where T3 Arabic outcomes were predicted by T2 Arabic environment factors, in addition to the Length of schooling in Arabic. The rationale for conducting both concurrent and longitudinal models for T3 outcomes was as follows: The effects of factors could be spread over time rather than concurrently; factors that emerge in both types of models could be considered relatively robust, and, if T2 factors predict T3 outcomes, this could give insights into the directionality of the association. T1 was not included in the modelling due to overlap with existing studies and because there is a long gap between T1 predictors and T3 outcomes. For both concurrent and longitudinal models, all fixed effects were numerical and were scaled and centered. The random effects for all four models included one intercept for Item and one for Participant, nested within Family. Initial models were simplified to the most parsimonious model possible, which was determined through backwards selection of fixed effects, using Log-likelihood ratio tests to ensure that pruning of the model did not result in a significant loss of fit. In some cases, this resulted in the inclusion of fixed effects showing only trends toward significance (p = 0.059–0.10) in the most parsimonious model. A summary of the results of the four models is in Table 2.
Summary of Factors Contributing to the Arabic SRT and Vocabulary Test in each Optimal Model
Note: Gold cells mean the factor is negatively associated with the outcome; blue cells indicate a positive association.
The most parsimonious concurrent model predicting Arabic SRT performance at T3 included Language use with siblings, Length of Arabic schooling, Speaking/listening in Arabic, Extracurriculars in Arabic, and Interactions with friends in Arabic (Model output is in Supplementary Appendix SI). Specifically, longer schooling in Arabic in Syria pre-migration (p = 0.003), higher frequency of Arabic speaking/listening activities (p = 0.068 [trend only]), and more frequent interaction in Arabic with friends (p = 0.015) were all positively associated with superior performance on the SRT. On the other hand, the model found that the more English that was used between participants and siblings, the lower the performance on the SRT (p = 0.028). Unexpectedly, the model found that a higher frequency of extracurricular activities (Koran classes) in Arabic was negatively associated with SRT performance (p = 0.021); we return to this anomalous result in the Discussion. For a longitudinal perspective, we modelled the SRT scores at T3 with T2 HL environment predictors, as well as the Length of Arabic schooling in the model. In the optimal model (Supplementary Appendix SI), two predictors showed a positive association with the outcome variable: Length of Arabic schooling (p = 0.006) and Hours of Arabic extracurricular activities at T2 (p = 0.098 [trend only]). On the other hand, Language use with siblings at T2 (p = 0.014) had a significant and negative association with the outcome variable.
The model predicting performance on the Arabic vocabulary test concurrently found two significant predictors: Length of Arabic Schooling and Language use with siblings. Specifically, participants with more months of Arabic schooling pre-migration (p < 0.001) performed better on the vocabulary test. On the other hand, using more English with siblings was associated with lower performance on the vocabulary test (p < 0.001; Model output is in Supplementary Appendix SJ). As we did for SRT, we modeled T3 vocabulary scores with T2 predictors. The optimal model (Supplementary Appendix SJ) showed that Extracurriculars in Arabic at T2 (p = 0.017) and Length of Arabic schooling (p < 0.001) were significantly and positively associated with the T3 outcomes. On the other hand, Language use with siblings at T2 was a negative predictor (p = 0.002).
8. Discussion
This study consisted of a prospective examination of the Arabic-HL development and the changing language environment of school-age, first-generation Syrian refugee children, who differed in their age at arrival but were resettled in Canada at the same time. Our objectives were (1) to determine if a shift from Arabic-HL to English-SL dominance emerged for language proficiency and frequency of language use over time, and how this was modulated by AOA; and (2) which language environment factors predicted individual variance in Arabic-HL outcomes at the final time period.
8.1. Shift from Arabic-HL to English-SL dominance in proficiency over time
With respect to developmental trajectories in their Arabic-HL, children showed growth on both the SRT and vocabulary test from T1 to T2, i.e., during their first three years of residency in Canada, regardless of AOA. Modest growth in Arabic word reading and reading comprehension was also found for a subset of this participant sample in their first three years in Canada (Al Janaideh et al., Reference Al Janaideh, Tibi, Gottardo, Paradis and Chen2023). However, by T3, we found evidence of attrition in SRT performance and a plateau in vocabulary performance, after four-and-a-half years in Canada. It is possible that this difference between the tasks could be due to expressive tasks being more challenging than receptive ones. For the SRT, a crossover in superior performance from Arabic to English – a shift in dominance from HL to SL – was evident for the younger AOA group at T3, and a trend in this direction emerged for the older AOA group. These developmental patterns for Arabic SRT and vocabulary contrasted with steady growth in children’s English across all time periods, AOA groups and tasks. In sum, our data revealed that these first-generation, school-age heritage bilinguals displayed a weakening of their Arabic-HL proficiency in favour of the English-SL during their first five years in Canada in terms of dominance, attrition or plateau in development. Thus, unlike Oppenheim et al. (Reference Oppenheim, Griffin, Peña and Bedore2020), we did not find evidence that growth in the HL occurred even with the dominant language shift.
While changes in developmental trajectories for Arabic over time were obtained regardless of AOA, our analyses showed that arriving in Canada at an older age was associated with stronger HL skills across all time periods. These findings provide further support for the possible protective effect of delayed onset of L2 acquisition in heritage bilinguals (Montrul, Reference Montrul2016, Reference Montrul2023), and are in line with much previous research showing an older onset of L2 advantage for the HL in second-generation bilinguals of all ages (see Age of Arrival in Heritage Language Development).
8.2. Shift from Arabic-HL to English-SL dominance in language environment over time
For our investigation into language environments, we examined the relative use of Arabic versus English at home among family members, and the frequency within an average week that children engaged in various language-rich activities in their Arabic and English. With respect to the former, Arabic remained the dominant language in interactions between children and their parents, across all time periods. The relatively stable and strong Arabic dominance in interactions with parents could be due, at least partly, to low levels of parental English proficiency (Paradis et al., Reference Paradis, Soto-Corominas, Chen and Gottardo2020). By contrast, language use in interactions between children and their siblings displayed a steady increase in English use overtime, regardless of sibling age. By T3, the average rating was between 3.5 and 4.5 on our scale, meaning both languages were used equally, or English was used more often amongst siblings. Finally, these patterns in language use at home were not different between AOA groups. Our findings are similar to those of G. Jia and Aaronson (Reference Jia and Aaronson2003) and Oppenheim et al. (Reference Oppenheim, Griffin, Peña and Bedore2020), who also found a gradual increase in the use of English-SL over time among Spanish-HL and Mandarin-HL school-age children in the United States. Therefore, on the one hand, the SL had not taken over as the dominant language of interaction among all family members after nearly five years in Canada, and the HL was holding its ground. On the other hand, the T3 results for language choice in interaction among siblings might lead to a situation common among adult HL speakers where parents end up being the primary or only source of interaction in the HL (Montrul, Reference Montrul2023).
Turning to engagement in language richness activities, following Daskalaki et al. (Reference Daskalaki, Soto-Corominas, Barisé, Paradis, Chen and Gottardo2023), we examined four different activities separately. We begin by comparing reading/writing activities, e.g., books, websites, messaging, homework, and speaking/listening activities, e.g., TV, streaming, music, and video chatting. Children’s engagement in reading/writing activities in Arabic was infrequent and stable across time and contrasted with the more frequent rates of engagement and growth in English reading/writing activities over time. Older AOA participants engaged more in reading/writing activities than their younger AOA peers, regardless of language, which is likely due to age-related differences in cognitive skills and school grades. The infrequent engagement in HL reading/writing, with no advantage for older AOA, differs from the findings of G. Jia and Aaronson (Reference Jia and Aaronson2003), and likely results from the interruptions in schooling experienced by many children in our sample, as well as arrival in Canada just before school entry age for our youngest participants. The patterns of engagement in speaking/listening activities over time were different from those of reading/writing activities. First, children, regardless of AOA, engaged in Arabic speaking/listening activities more frequently in an average week than in Arabic reading/writing activities, with ratings often in the 3–4 range, meaning between 5 and 20 h per week, regular-to-often. Furthermore, contrary to reading/writing activities, children were more frequently engaged in Arabic speaking/listening activities than English ones at T1. However, this shifted to more frequent engagement in English activities by T3 across AOA groups.
Regarding activities outside the home, we found that children were using more Arabic than English with their friends at T1, but this reversed to more use of English at T2 and T3, regardless of AOA. Nevertheless, the mean scores for the use of Arabic with friends were from 2 to 3 throughout, so Arabic was used sometimes-to-regularly with friends, even if surpassed by English overtime. In contrast, children did not participate frequently in extra-curricular activities in either language (ratings between 1 and 2, between 0 and 5 h), which could be expected because, as newly arrived refugee families with many children and living on social assistance (Paradis et al., Reference Paradis, Soto-Corominas, Vitroulis, Al Janaideh, Chen, Gottardo, Jenkins and Georgiades2022), it is likely that funds for such activities would be scarce. Moreover, for the vast majority of children, extra-curricular activities in Arabic consisted only of religious courses, often referred to as “Koran class”, at the Mosque. According to qualitative comments from parents, such classes varied in activities carried out, the extent to which different varieties of Arabic were used, e.g., classical Arabic, MSA, or spoken varieties, and whether literacy skill-building took place. In English, by contrast, there was no dominant extra-curricular activity. The decrease in English activities to near zero at T3 was due to pandemic-related closures, reducing opportunities for attending low- or no-cost sports and clubs in English at school.
In conclusion, language use at home among siblings, frequency of listening-speaking activities and use of Arabic with friends showed shifts from HL to SL dominance over time. Thus, even for first-generation school-age children, the HL environment can be vulnerable and prone to change in favor of the SL, and older AOA is not protective against these changes, unlike for HL proficiency. For a “glass-half-full” interpretation, it should be noted that, even after almost five years in Canada, participants were still using their Arabic-HL regularly with their siblings and friends and were engaging regularly in speaking/listening activities in Arabic. A mix of the HL and SL in the language environment during the school years, with a general downward trend for the HL, is consistent with Carreira and Kagan’s (Reference Carreira and Kagan2011) retrospective study on adult heritage bilinguals, and thus, could be the most realistic expectation for migrant children.
8.3. Sources of individual differences in the Arabic-HL after four-and-a-half years in Canada
Our final analyses were aimed at understanding the role of the dynamic language environment factors, as well as the length of Arabic schooling pre-migration, in predicting individual differences in Arabic-HL outcomes at T3. Concurrent and longitudinal analyses were employed to infer robustness and potential directionality in associations.
Language use among siblings emerged consistently across the concurrent and longitudinal models for both the SRT and vocabulary test such that, more use of English (=less use of Arabic) with siblings was predictive of lower performance on Arabic tasks. We interpret these findings to mean that language choice among siblings could be a risk factor for HL attrition. While language use and language proficiency can mutually influence each other, the longitudinal association suggests the possibility of directionality, where more frequent use of English among siblings could be eroding participants’ Arabic skills over time. The strength of the sibling language use variable could be influenced by the size of the families in our sample. With most families having 3–6 children (Paradis et al., Reference Paradis, Soto-Corominas, Vitroulis, Al Janaideh, Chen, Gottardo, Jenkins and Georgiades2022), sibling interlocutors outnumbered parent interlocutors in the home. While language use with siblings showed robust associations with variation in Arabic outcomes, language use with parents did not, and this is likely due to all parents using Arabic almost exclusively with their children. Furthermore, the frequency of Arabic speaking/listening activities, and speaking Arabic with friends were only significant in the concurrent model for SRT, so these language richness components were not as robustly associated with HL outcomes as sibling language use in our study. Therefore, our findings support Czapka et al. (Reference Czapka, Topaj and Gagarina2021) regarding the importance of sibling language use in HL development and maintenance.
Another factor that predicted stronger performance on both tasks consistently across concurrent and longitudinal models was the length of Arabic schooling pre-migration. Note that mutual influence is unlikely here because Arabic abilities at T3 could not possibly influence the length of Arabic schooling as this factor was constant overtime. The influence of some schooling in the HL, 5 years or more in the past, on current HL abilities, testifies to the significance and tenacity of this factor. Our results suggest that, for refugee children who often have limited and interrupted education in their HL, having at least some education pre-migration could be a protective factor for HL maintenance. Overall, our findings are consistent with studies of other bilingual populations that have found that HL educational experiences contribute to building stronger HL abilities (Bayram et al., Reference Bayram, Rothman, Iverson, Kupisch, Miller, Puig-Mayenco and Westergaard2017; R. Jia & Paradis, Reference Jia and Paradis2015; Kupisch & Rothman, Reference Kupisch and Rothman2018; Rodina et al., Reference Rodina, Kupisch, Meir, Mitrofanova, Urek and Westergaard2020; Torregrossa et al., Reference Torregrossa, Flores and Rinke2023b).
In the longitudinal analyses for both tasks, the frequency of participation in extra-curricular activities in Arabic at T2 was positively associated with HL outcomes at T3. On the one hand, this is a surprising result, as children did not participate frequently in any extra-curricular activities (see Figure 4). On the other hand, the frequency of these activities was positively correlated with both the length of schooling in Arabic (r = 0.236; p < 0.001) and the frequency of interaction in Arabic with friends at T3 (r = 0.313; p < 0.001), suggesting a common underlying construct and some overlap among these variables (see Supplementary Figure S1). Additionally, it is possible that attending Koran classes at the Mosque conferred advantages for HL development beyond quantity/quality of Arabic input, for example, contributing to cultural identity and positive attitudes towards the heritage language and culture. Paradoxically, in the concurrent model for SRT, participation in extracurriculars in Arabic had a negative effect on scores and was not a significant factor in the concurrent model of vocabulary. The anomalous and non-effect of participation in extracurriculars in Arabic at T3 is difficult to explain. One reason could be differences in the quality of the actual activities at T3 due to the pandemic. It could also be some idiosyncracies in individual performance on the SRT among participants who attended Koran classes in our T3 dataset.
8.4. Role of the COVID-19 pandemic
The third wave of data collection for this study took place during the pandemic, which meant that there were intermittent school closures and remote learning. Therefore, children would have been spending less time in an English-SL environment and more time at home in an Arabic-HL environment than at T1 and T2. This could have, in principle, been protective against HL attrition and/or detrimental to L2 development. However, our data display the opposite pattern. Increases in some English richness activities, English use among siblings and English proficiency were evident from T2 to T3, along with decreases in Arabic use at home with siblings, some Arabic richness activities and Arabic proficiency. In sum, the rise of the English-SL in both environment and proficiency appeared to proceed regardless of school closures and remote learning for these heritage bilinguals.
9. Limitations and conclusions
There are several limitations to this study. First, due to circumstances beyond our control, we had to switch from in-person to online testing at T3. While we believe our data to be sound generally (see Data collection), it is never optimal to switch testing methods in a longitudinal study. Another limitation was that the vocabulary tests could not be compared directly, thus limiting the ability to detect dominant language shift to just one task. Regarding language richness factors, we relied on parent reports, rather than youth self-report. Given the age of the children at T3 especially, youth self-report could have been a more sensitive measure. Finally, first-generation refugee children have some differences from other migrant children which impact language development, such as interrupted education and other adversity factors (Paradis et al., Reference Paradis, Soto-Corominas, Vitroulis, Al Janaideh, Chen, Gottardo, Jenkins and Georgiades2022), this signals caution in the generalization of findings to all heritage bilinguals.
Limitations notwithstanding, this is one of the very few prospective studies on HL development in first-generation bilinguals, examining both language proficiency and frequency of use as well as individual differences in HL outcomes. The children in this study exhibited some growth and reasonable maintenance of their Arabic-HL during their first five years in the host country, which could be expected given they were first-generation migrants. Nevertheless, there were signs that a shift to English-SL dominance in both proficiency and language use, and attrition of Arabic-HL skills were taking place, particularly for younger arrivals, by the end of the study. Moreover, as HL maintenance in the long term is dependent on HL use at home and in other activities, further increases in the use of the English-SL could exacerbate the trends revealed in this study. In short, even first-generation bilingual children could be vulnerable to HL attrition. Because, in contrast to AOA, language environment factors are malleable, our results strongly suggest that families be vigilant about the quantity and quality of HL experiences at home if children’s HL maintenance in the longer term is a goal.
Supplementary material
The supplementary material for this article can be found at http://doi.org/10.1017/S0305000924000679.
Acknowledgments
We want to thank the Syrian-Canadian families for the time and effort it took to participate in this study. We would also like to thank the more than 20 research assistants across three cities who collected the data. This research was funded through a Partnership Grant from the Social Sciences and Humanities Research Council of Canada, for which we are grateful (SSHRC PG: Ungar [PI], Paradis and Chen [Co-Is]).
Competing interest
The authors declare none.
Open access
