1. Introduction
Language input plays a vital role in bilingual lexical acquisition and its notorious heterogeneity. It is established, for instance, that lexical proficiency in each language is likely to reflect the relative amount of input children receive in both languages, besides qualitative factors such as varying language use according to setting and interlocutor language proficiency (see Armon-Lotem & Meir, Reference Armon-Lotem, Meir, De Houwer and Ortega2019, for a review). Individual differences among bilingual children are more pronounced than in monolinguals because of bilingual environments’ intrinsic variability (Hoff, Reference Hoff2020). The amount of exposure to each language, the age at which exposure began, and the number of interlocutors speaking each language (e.g., Hoff et al., Reference Hoff, Welsh, Place, Ribot, Grüter and Paradis2014; O’Toole et al., Reference O’Toole, Gatt, Hickey, Haman, Miękisz, Armon-Lotem, Rinker, Ohana, Santos and Kern2017) are only a few of the sources of variation documented in bilingual lexical development, even when children share the same input language pair (Hoff & Core, Reference Hoff and Core2013). Typological distance between languages adds to the individual differences in bilingual children’s lexical skills, as it determines the extent of cross-language knowledge transfer possible (Blom et al., Reference Blom, Boerma, Bosma, Cornips, van den Heuij and Timmermeister2020). Further, the research literature shows mixed-language input as influencing children’s bilingual acquisition in various ways (Ruan et al., Reference Ruan, Byers-Heinlein, Orena and Polka2023), possibly also because of differences between bilingual communities in the borrowing and code-switching patterns employed (Kremin et al., Reference Kremin, Alves, Orena, Polka and Byers-Heinlein2022). Indeed, bilingual input is shaped by the social context in which it is embedded (Hoff, Reference Hoff2020). Distal factors such as family socioeconomic background, attitudes and policies towards bilingualism, as well as languages’ majority/minority status are among the broader factors that compound variability in children’s bilingual outcomes, including their lexical skills (Paradis & Grüter, Reference Paradis, Grüter, Grüter and Paradis2014; Thordardottir, Reference Thordardottir, Grüter and Paradis2014). A robust relationship between socioeconomic status (SES) and lexical acquisition is documented for monolingual and bilingual children alike, with higher-SES children consistently performing better on lexical tasks than their lower-SES peers (e.g., Fernald et al., Reference Fernald, Marchman and Weisleder2013; Hoff, Reference Hoff2003; Montanari et al., Reference Montanari, Mayr and Subrahmanyam2020; O’Toole et al., Reference O’Toole, Gatt, Hickey, Haman, Miękisz, Armon-Lotem, Rinker, Ohana, Santos and Kern2017; Rowe, Reference Rowe2012). The effects of SES on children’s lexical skills are reported to be similar to those of caregiver input.
1.1. Language input links SES to child’s lexical ability
Specific properties of caregiver input have been found to mediate the effects of children’s more distal socioeconomic background (e.g., Hart & Risley, Reference Hart and Risley1995; Huttenlocher et al., Reference Huttenlocher, Waterfall, Vasilyeva, Vevea and Hedges2010; Stein et al., Reference Stein, Menti and Rosemberg2021). The total number of words, or word tokens, and the range of different words, or types, used by caregivers are employed as indices of input quantity and quality, respectively. In their landmark study, Hart and Risley (Reference Hart and Risley1995) reported family SES to be significantly related to the amount of words per hour that children aged 10 to 36 months were spoken to by their parents, with children from high-SES families exposed to three times the amount of words received by their low-SES counterparts. Similar findings have been reported for languages other than English. For instance, more word tokens and word types were received by children in mid-SES Argentinean households when compared to low-SES ones (Stein et al., Reference Stein, Menti and Rosemberg2021). In turn, children’s lexical abilities are influenced by the total number of words, or input quantity, and the range of different words, or qualitative lexical diversity, received from their caregivers. There is widespread consensus that the more abundant and diverse the language input is, the better the children’s lexical abilities will be. For example, Hoff (Reference Hoff2003) reported that the total number of utterances, word tokens, word types, and topic-continuing replies used in maternal input were related to the vocabulary abilities of English-speaking 16- to 31-month-olds. In Rowe’s (Reference Rowe2012) longitudinal study of parent–child dyads having English as the primary language, 30-month-olds whose parents used more diverse vocabulary achieved higher receptive vocabulary scores after 1 year than peers exposed to less rich vocabulary. Similar findings have been reported for 18-month-olds learning Mexican Spanish from parents having limited English proficiency, most of whom were recent migrants to the United States. Spanish was the only language reportedly spoken in these primarily low-income homes. Children whose mothers used more Spanish word tokens and types during free play experienced larger increases in receptive vocabulary and grew faster at word recognition in just 6 months (Hurtado et al., Reference Hurtado, Marchman and Fernald2008). Anderson et al.’s (Reference Anderson, Graham, Prime, Jenkins and Madigan2021) meta-analysis compared associations between parental input and child language reported in observational studies. Despite input quantity and quality being highly correlated, the effect of qualitative measures on English language development was found to be more robust. Further, input quality effects on children’s outcomes seem related to the language skills already available to children, appearing more relevant to toddlers and pre-schoolers rather than to infants at the earliest stages of monolingual language acquisition (Golinkoff et al., Reference Golinkoff, Hoff, Rowe, Tamis-LeMonda and Hirsh-Pasek2019; Rowe & Snow, Reference Rowe and Snow2020).
In low-SES environments, caregiver input tends to have diminished quantity and quality due to the adverse effects that increased economic stress may have on parenting skills (Perkins et al., Reference Perkins, Finegood and Swain2013). In fact, low-SES families are likely to invest more time and effort in acquiring resources related to basic needs when compared to more economically and financially stable caregivers who can afford to purchase more expensive literacy and educational materials (Conger & Donnellan, Reference Conger and Donnellan2007). In itself, SES is a multi-factorial construct, incorporating an individual’s educational success, economic position, and occupation in relation to other individuals in society (Manstead, Reference Manstead2018). Studies investigating socioeconomic effects vary in the manner by which they classify participants within each socioeconomic stratum. Some studies measure SES through a single factor, with maternal education being a preferred index of SES, based on the fact that mothers are the primary caregiver in the majority of households (Hoff, Reference Hoff2003). Other studies employ more complex operationalisations of SES. For example, Roy et al. (Reference Roy, Chiat and Dodd2014) stratified children into different SES groups using the index of multiple deprivation, which considers family units’ income, employment, education, health, crime, living environment, as well as barriers to housing and services. Effects on both receptive and expressive vocabulary skills of children aged 3;06–5 years were reported. Indices of SES that include multiple dimensions are more informative than simple, unitary measures and are likely to be better predictors of children’s language proficiency (De Cat, Reference De Cat2021; Gatt et al., Reference Gatt, Baldacchino and Dodd2020).
As with monolingual acquisition research, methodological inconsistencies in the operationalisation of SES are evident in studies of bilingual acquisition. Moreover, the relationship between SES, language input characteristics, and lexical outcomes is much more complex in bilingual children than in monolinguals. Not only does SES manifest a robust influence on bilingual lexical proficiency; it also interacts with children’s bilingual experience in specific ways. Substantial research has focused on children from low-SES immigrant families who grow up in minority language homes and learn the majority language through schooling. For example, several studies on minority Spanish-speaking children in the United States have shown that low SES affects L2 lexical skills unfavourably, with no effect on L1 proficiency (e.g., Buac et al., Reference Buac, Gross and Kaushanskaya2014; Hoff et al., Reference Hoff, Giguere, Quinn and Lauro2018; Montanari et al., Reference Montanari, Mayr and Subrahmanyam2020). Yet, studies of Spanish–English bilinguals based in South Florida have shown that high SES is also possible in minority language groups (e.g., Hoff et al., Reference Hoff, Core, Place, Rumiche, Señor and Parra2012, Reference Hoff, Giguere, Quinn and Lauro2018). In contexts where bilingualism is stable and not associated with a particular SES group, bilingual proficiency in children learning L2 English alongside a minority home language can be investigated across a range of SES levels. This is also the case for Welsh children acquiring English, the majority language, alongside Welsh, a minority language (Gathercole, Reference Gathercole, Nicoladis and Montanari2018).
Malta’s bilingual context lends itself well to studies on bilingual language acquisition investigating SES effects. Maltese children grow up with Maltese and English, both official and majority languages, and are all bilingual to varying degrees, regardless of their SES. This means that their bilingual acquisition can be investigated with minimal risk of SES and bilingual input effects being confounded (Gatt et al., Reference Gatt, Baldacchino and Dodd2020). The two input languages of Maltese children are also typologically very distant, with Maltese being Semitic in origin and English having Germanic roots. Studies of child language situated in countries where bilingualism in two majority languages is the norm are uncommon. Research in normative bilingual settings can make an important contribution to our theoretical understanding of bilingual language acquisition (Montanari & Nicoladis, Reference Montanari, Nicoladis, Nicoladis and Montanari2016).
While English enjoys considerable prestige in Malta (Formosa & Little, Reference Formosa and Little2023), the preferred spoken language for 92.2% of Maltese individuals aged 5 years and over is Maltese, the national language (National Statistics Office, Malta, 2024). Accordingly, most young children are expected to receive Maltese-dominant exposure in their homes. Smaller segments of the Maltese childhood population are raised with simultaneous exposure to Maltese and English or with a predominance of English. The use of English as a home language tends to be associated with higher social status, and children may be reluctant to use Maltese in English-speaking Maltese families (Formosa & Little, Reference Formosa and Little2023). In contrast, English features prominently at the Maltese-dominant end of the exposure continuum. Gatt and Dodd (Reference Gatt and Dodd2019) described the language input of children in Maltese-dominant families as typically consisting of Maltese syntactic structures embedding English code switching and lexical borrowing, with the latter including core and functional borrowing. Core borrowings are words often preferred over available Maltese equivalents in adult conversational exchanges. The term “functional borrowings” was coined for words often preferred over their Maltese equivalents by Maltese-speaking adults when addressing young children, but not in adult conversations. Gatt et al. (Reference Gatt, Grech and Dodd2016) reported that Maltese-dominant mothers exposed their toddlers to English regardless of their educational background, with those having a higher level of education possibly modelling English that was less fragmented and more cohesive. Unless regular exposure to both Maltese and English takes place simultaneously in the home, children would go on learn their second language (Maltese or English) through schooling (Mifsud & Vella, Reference Mifsud and Vella2020). A recent study on Maltese-dominant 4-year-olds investigated the effects of SES on children’s Maltese and English vocabulary abilities and found a composite measure of SES, incorporating both maternal and paternal occupation and education, to be the strongest predictor of lexical performance when compared to individual measures (Gatt et al., Reference Gatt, Baldacchino and Dodd2020). The present study goes a step further, investigating the bilingual lexical skills of three-year-old children, having different socioeconomic backgrounds and raised in Maltese-dominant homes, in relation to caregiver input.
1.2. Wanted: absolute measures of bilingual input
As in monolingual acquisition, the influence of SES on bilingual vocabulary performance is likely channelled through its impact on bilingual input properties. Notably, however, much of what we know about input effects on bilingual language acquisition draws on relative rather than absolute measures (Armon-Lotem & Meir, Reference Armon-Lotem, Meir, De Houwer and Ortega2019; Paradis & Grüter, Reference Paradis, Grüter, Grüter and Paradis2014). Questionnaires and interviews are often used to elicit caregiver estimates of the relative amounts of input received by children in each language. Some studies employed daylong recordings of children’s language input to obtain direct measures of the number of words received. Marchman et al. (Reference Marchman, Martínez, Hurtado, Grüter and Fernald2017) reported that the counts of English and Spanish words directed to three-year-olds, based on naturalistic audio recordings, were significantly associated with the children’s outcomes in each language. Although absolute numbers of words used by children’s parents corresponded with estimated input amounts, direct measures showed more robust associations with children’s language outcomes. Using a similar set-up, Orena et al. (Reference Orena, Byers-Heinlein and Polka2020) found that the absolute numbers of words received by 10-month-old infants in French–English bilingual families varied considerably in children reportedly exposed to similar proportions of exposure. In contrast, Verhoeven et al. (Reference Verhoeven, van Witteloostuijn, Oudgenoeg-Paz and Blom2024) did not find differences in the reliability of reported and observational measures of input quantity for 3- to 5-year-old Dutch L2 learners. Together, these findings suggest that absolute measures are more objective indices of input quantity than relative ones, although preference of one method over the other may be dictated by the circumstances of data collection. Nonetheless, the effects of caregiver input quality on bilingual children’s vocabularies remain under-researched (Huang & Kuo, Reference Huang and Kuo2020). Studies investigating effects of direct measures of caregiver input quality on bilingual children’s language skills are noticeably scarce (Gámez et al., Reference Gámez, Palermo, Perry and Galindo2023). This could be because determining the number of different words in naturalistic samples requires more effort than counting all words indiscriminately without differentiating repeated ones. In a sample of low-income immigrant families in the United States, children’s vocabulary growth in English and Spanish between 2 and 5 years was associated with changes in maternal input quality in both languages, as measured by word types, during book sharing (Tamis-LeMonda et al., Reference Tamis-LeMonda, Song, Luo, Kuchirko, Kahana-Kalman, Yoshikawa and Raufman2014). More recently, Gámez et al. (Reference Gámez, Palermo, Perry and Galindo2023) tallied the Spanish and English word types used by caregivers with their 18-month-old infants during 75-minute recordings of naturalistic interactions. Caregivers’ lexical diversity in Spanish positively predicted children’s Spanish and total expressive vocabularies at 24 months, while English type counts positively influenced children’s English expressive vocabulary size.
Thorough examination of children’s bilingual vocabularies needs to acknowledge the operation of environmental and input factors shaping children’s bilingual experience, besides child-internal differences in language learning ability. Our current understanding of how direct measures of caregiver input affect bilingual vocabulary acquisition across different SES levels often relies on the extrapolation of findings for monolingual children. Studies addressing the effects of absolute measures of caregiver input on the bilingual lexical skills of children having diverse SES backgrounds are possibly non-existent. This lack of research stems from an amalgamation of methodological obstacles, specifically the challenge of disentangling the effects of SES and bilingual experience, as well as the laborious task of measuring bilingual caregiver input directly, particularly when input quality is the focus. As a result, our current understanding of this under-researched aspect draws on two separate strands of research. The influence of SES on children’s bilingual lexical outcomes has been the topic of several investigations spanning various language pairs, in a range of bilingual settings. Findings generally show better L2 lexical outcomes in the presence of higher-SES levels (e.g., Hoff et al., Reference Hoff, Core, Place, Rumiche, Señor and Parra2012, Reference Hoff, Giguere, Quinn and Lauro2018; O’Toole et al., Reference O’Toole, Gatt, Hickey, Haman, Miękisz, Armon-Lotem, Rinker, Ohana, Santos and Kern2017). The few studies addressing bilingual vocabulary skills in relation to direct measures of bilingual input indicate that the latter are positively associated with children’s vocabulary size in each language (e.g., Gámez et al., Reference Gámez, Palermo, Perry and Galindo2023). To our knowledge, research investigating the effects of SES and absolute measures of input on children’s bilingual vocabulary skills is non-existent. In attempting to address this specific gap in the empirical literature, the present study analysed three-year-olds’ Maltese and English lexical skills as a function of primary caregiver input and socioeconomic background. The following research questions were addressed:
-
1. Is there a relationship between SES, as measured through a composite measure of parental occupation and education, and the receptive and expressive lexical skills of Maltese-dominant three-year-olds, measured in Maltese and English?
-
2. Does primary caregiver language input predict children’s bilingual lexical skills?
-
3. How do children’s bilingual lexical skills vary when both SES and caregiver input are considered?
-
4. Is there an association between SES, single-language measures of Maltese and English word types in caregiver input, and children’s lexical skills in each language?
2. Materials and methods
2.1. Participants
Participants were 38 children (11 males) aged 3;04–3;08 years (M = 3.29, SD = 1.45), along with their respective primary caregivers. The limited age range controlled for developmental variability, reducing its potential influence on lexical performance scores. The caregivers were 36 mothers, a father, and a grandmother. Children were recruited from public pre-schools distributed randomly across Malta. Language exposure is said to be divided relatively equally between Maltese and English in public education in Malta (Pace & Borg, Reference Pace and Borg2017). The participants were required to be sequential bilinguals who came from predominantly Maltese-speaking households, reflecting the Maltese dominance prevalent on a national scale (National Statistics Office, Malta, 2021). According to the main caregivers, all children received Maltese-dominant language exposure, regardless of SES background. Thus, the children’s regular exposure to English was expected to commence upon pre-school entry, making English their second language. Table 1 shows the frequency distribution of girls and boys relative to SES group and geographical region of the pre-school attended.
Table 1. Frequency of children within each SES category according to geographical region
2.1.1. Socioeconomic status (SES)
Points in the range of 0–4 were assigned to paternal and maternal education and occupation, as reported by primary caregivers, and a composite SES score was derived. Compound measures of SES are known to account for more variability in children’s language scores than unitary measures (De Cat, Reference De Cat2021; Gatt et al., Reference Gatt, Baldacchino and Dodd2020). Table 2 shows the allocation of points for educational level and occupational status category. Occupations were coded according to Ganzeboom et al.’s (Reference Ganzeboom, Graaf and Treiman1992) International Socio-Economic Index of Occupational Status, and cut-off points between the occupation categories were derived by dividing the total range of scores (10–90) into five more or less equal groups. Points allocated to the educational and occupational level of each parent were then summed up. The resulting compound SES scores, one for each child participant, were divided into low (N = 10, range = 2–5), medium (N = 17, range = 6–9) and high SES groups (N = 11, range = 10–15).
Table 2. Points assigned for maternal and paternal education and occupation levels
2.2. Methods and tools
Caregivers reported on their children’s language-related developmental milestones, relative amounts of Maltese and English input received within and outside of the home, the amount of time each caregiver spent interacting with the child on a daily basis, as well as maternal and paternal education and occupation levels, by completing a questionnaire adapted from the Questionnaire for Parents of Bilingual Children (COST IS0804, 2011).
Direct assessment provided measures of caregiver input and children’s lexical skills. Child-directed language used by the caregivers was sampled in the children’s homes for a period of 20 minutes. Each adult–child dyad was provided with a standard set of “role play” toys, which consisted of a tea set, a cooking set, cars, tools, a sandwich preparation game, a teddy bear, and a tiger puppet. By making use of the same set of play materials, each adult–child dyad had similar opportunities to interact during free play. Each dyadic interaction was audio-recorded and later transcribed orthographically. Four randomly selected audio recordings (10%) were checked for inter-transcriber reliability, which amounted to 99.1%. A pilot study established that the optimal sample duration and time segment for type and token analysis was the middle 10 minutes of each interaction. Productions that were ambiguous or were perceived to carry a vague meaning at the lexical level on account of their simple phonological form, such as “hmm” or “aa,” were not transcribed. Type and token counts were derived manually from the transcribed dyadic interactions and employed as measures of input quality and quantity (see e.g., Hurtado et al., Reference Hurtado, Marchman and Fernald2008; Rowe, Reference Rowe2008).
Children’s lexical abilities were assessed using Maltese and English versions of a Receptive Picture Name Judgement Task and a Picture Naming Task. These instruments are not norm-referenced, but their construction and content have been described in detail, and they have previously yielded valid and reliable documentation of Maltese children’s lexical skills (see Gatt et al., Reference Gatt, Baldacchino and Dodd2020; Gatt & Dodd, Reference Gatt and Dodd2019). The four tasks were each composed of 40 test items. The receptive tasks tapped into the children’s ability to judge the accuracy of phonological representations for familiar words. By tapping into children’s integrity of phonological representations, which is an aspect of phonological processing ability, the tasks differed from traditional receptive vocabulary tasks that assess receptive vocabulary through picture identification (Gatt & Dodd, Reference Gatt and Dodd2019). Each task consisted of 40 stimuli, which were presented to the participants as labels to coloured photographs presented individually on a laptop screen by the second author. Nineteen of the stimuli in the English task and 21 of the Maltese stimuli were phonological foils that comprised changes in place, manner, and voicing of production, for example, “pocolate” in place of “chocolate”; “par” for “car”; “vish” for “fish.” Children were asked “Is this a/an….?” and were required to answer with “yes” if the word was phonologically accurate and “no” if it was not. Maltese and English picture naming tasks were used to assess lexical access and the accuracy of the child participants’ semantic representations. Each expressive task was meant to elicit the production of 40 common nouns in the language of testing. Precisely 40%, or 16 photographs, showed concrete nouns that were common to both tests, allowing the analysis of translation equivalents, if necessary. Stimuli common to both language assessments were depicted through different images, to minimise interference between the test languages (Gatt & Dodd, Reference Gatt and Dodd2019).
2.3. Data gathering procedure
Permission was sought and granted from the Directorate for Research, Lifelong Learning and Employability within the Ministry of Education and Employment to seek potential participants from a variety of state schools. Approval to carry out data collection was granted by the Faculty of Health Sciences Research Ethics Committee at the University of Malta (Ref. 121/2017). Schools with potential child participants were identified. Caregiver questionnaires and consent forms were made available to potential adult participants by the respective schools. The adults who consented to participating were then contacted and an appointment for a home visit set up. The Maltese and English vocabulary tasks and sampling procedure were conducted on the same day by the second author, with the full assessment procedure lasting approximately 1 hour. The lexical tasks were carried out first, with a short game separating the Maltese and English versions. For each language, receptive judgement preceded picture naming. Although counterbalancing of the languages of assessment was originally planned, piloting indicated that some children lacked confidence in starting assessment in their less preferred language. To avoid impacting children’s performance negatively, children were asked to indicate the language they preferred to start with. Twelve children chose to start with the English tasks.
2.4. Data coding
2.4.1. Coding of caregiver input
Caregiver input quality was represented in terms of lexical diversity by tallying the different Maltese words (types), the different English words, and the total number of different words (aggregated count of Maltese and English types) that each adult produced during play with their child. The total number of words used represented input quantity. Decisions pertaining to which word tokens should be realised by the same types and which words would best be separated into their various meaningful morphological components, when prefixes and suffixes were present, are outlined in Table A1 (Appendix). To facilitate extraction of types from tokens, the “A to Z function” was selected on Microsoft Excel so that alike words were listed after each other and a single word from a list of similar words could be identified as a type more easily. Once the various word forms (tokens) were sorted into their respective type groups, the total number of types and tokens within each language sample was obtained through a computerised word counting procedure.
In identifying words as Maltese or English types, criteria adopted in Gatt et al. (Reference Gatt, Grech and Dodd2016) were followed. Besides native Maltese words, Maltese type counts comprised established English borrowings that did not have a Maltese equivalent and that had been imported into Maltese out of necessity, for example, burger, stickers. The English words counted as Maltese amounted to 5.83% of the total Maltese type count. English type counts included core borrowings, or words often preferred over available Maltese equivalents in adult conversational exchanges, for example, baby, thank you, and functional borrowings, that is, words often preferred over Maltese equivalents by Maltese-speaking adults when addressing young children, but not in adult conversations, for example, yellow, sheep. Caregiver’s use of core and functional borrowing with their children may not have been regulated entirely by societal trends and might have reflected individual language choices driven by the circumstances of the specific adult–child dyad. Apart from borrowing, English word counts also considered instances of code switching. All English lexical items that were used despite a Maltese equivalent being available, that is, core and functional borrowings and code-switched elements, were therefore coded as English words. The range of English words sampled in adult input during play provided an index of the degree of language mixing employed by the caregiver with the child. English word types were considered as separate from their Maltese translation equivalents, if these were also sampled. Inter-rater agreement in the coding of all Maltese and English words in the sampled caregiver utterances amounted to 94.51% for English types and 96.17% for Maltese. Cognate terms having similar phonological and semantic representations that could not be clearly attributed to Maltese or English, for example, banana, blu/blue, were coded as “non-specific language words” and counted separately. They amounted to 2.22% of the total type count and were not analysed further.
2.4.2. Coding of children’s performance on the lexical tasks
Coding of children’s lexical task performance was based on the instructions for data coding and inputting accompanying the tasks (Gatt & Dodd, Reference Gatt and Dodd2019). On the receptive judgement tasks, children’s correct responses were assigned one point, while incorrect responses were given none. Incorrect responses included the following: (1) an unrelated or conflicting response (e.g., Adult: “Is this a hoy?” Child: “No, it’s a hoy”), (2) an inability to detect a phonological foil in a stimulus, (3) an inability to identify that a word contained no phonological foils or wrongly correcting a correct production using a different label altogether, for example, Adult: “Is this a dog?” Child: “No, chicken” (semantic error), and (4) a reply in the non-test language (language error). For the picture naming tasks, responses were assigned a point when they were semantically accurate and occurred in the language of testing. Semantically accurate responses in the non-test language were scored separately, but were not analysed in the present study. The maximum score possible on each of the four tasks was 40. A 100% inter-rater agreement score was obtained when a speech-language pathologist independently coded 10% of children’s responses with the assistance of the coding instructions accompanying the tasks.
3. Results
The results are organised according to the four research questions driving the study.
3.1. Research question 1: Is there a relationship between composite SES and children’s receptive and expressive lexical skills in Maltese and English?
Children’s lexical skills in Maltese and English were first analysed in relation to their SES (see Figure 1).
Figure 1. Children’s mean scores on receptive picture name judgement and picture naming in Maltese and English as a function of SES level (errors bars represent 95% confidence intervals).
Examination of children’s scores at each SES level showed children’s receptive judgement to be consistently better, on average, than their picture naming, for both Maltese and English (low SES: for Maltese t(9) = 13.989, for English t(9) = 6.161, p < .001; medium SES: for Maltese t(16) = 26.010, p < .001, for English t(16) = 9.875, p < .001; high SES: for Maltese t(10) = 4.949, p = < .001, for English (t(10) = 6.423, p < .001). Of note are the significantly higher mean scores for English picture naming, compared to Maltese, for children in the medium (t(16) = −4.108, p < .001) and high (t(10) = −3.298, p = .004) SES groups. Despite the children’s Maltese-dominant home environments, the gap between receptive judgement and picture naming was larger for Maltese than for English. With increasing SES, the improvement in picture naming performance was considerable for English, with multivariate regression analysis showing significant differences in mean scores between the low and medium SES groups (unstandardised coefficient B = −13.327, p = <.001, η2. = .312) and between the medium and high SES groups (B = −8.021, p = .010, η2 = .173). A significant SES effect also resulted for English receptive judgement, with the low SES group scores being, on average, 5.297 scale points lower than high-SES scores (p = <.001, η2 = .373). Thus, English receptive judgement varied markedly between low and high SES groups, and English picture naming even tapped into differences between low, medium, and high SES levels. In contrast, Maltese receptive judgement and picture naming only differed marginally between the three SES groups. Maltese receptive judgement mean scores were not significantly different from their English counterparts.
3.2. Research question 2: Does primary caregiver language input predict children’s bilingual lexical skills?
Children’s lexical performance was next examined in relation to caregiver input. First, descriptive statistics were computed for aggregate token and type counts, which considered Maltese and English collectively. These are shown in Table 3, together with means, standard deviations, and ranges for children’s performance on each lexical task. As high and significant correlations resulted between the type and token counts (r = .659, p < .001), subsequent analyses only considered word types to minimise multicollinearity effects. Multivariate regression analysis showed caregiver input quality to be significantly related to children’s performance on the English lexical tasks (receptive judgement: F(1, 36) = 5.05, p = .031, η2 = .123; picture naming: F(1, 36) = 17.67, p < .001, η2 = .329), but not on the Maltese ones. Type count effects seemed to be mirroring those of SES, with both variables predicting children’s English receptive judgement and naming performance and none of them bearing an influence on Maltese lexical skills. So, the next analyses sought to determine whether the association between SES and English lexical performance was mediated by the number of types in caregiver input.
Table 3. Means, standard deviations (SD), and ranges for type and token counts in children’s input and lexical task scores (N = 38)
3.3. Research question 3: How do children’s bilingual lexical skills vary when both SES and caregiver input are considered?
To address this question, three linear regression models were first fitted for English receptive judgement. SES significantly predicted children’s receptive scores (F(2,35) = 11.033, p < .001, η2 = .387) and the aggregate type count (F(2,35) = 8.023, p < .001, η2 = .314). The third model related children’s receptive judgement performance to both SES and aggregate type count. Here, the effect of SES on English receptive judgement was significant (F(2,34) = 7.354, p = .002, η2 = .302), but the type count effect was not (p = .801), indicating that caregivers’ input quality was not mediating the SES effects on English receptive judgement. Figure 2 summarises the results for the mediation model fitted for English picture naming. SES significantly predicted children’s picture naming scores (F(2,35) = 8.159, p = .001, η2 = .318) with effect (unstandardised coefficient) c = −13.327. SES also had a significant effect on caregiver type count (F(2,35) = 8.023, p = .001, η2 = .314), with effect a = −47.018. When English picture naming was examined in relation to SES and type counts, the SES effect did not remain significant (F(2,34) = 2.487, p = .098, η2 = .128; effect
$ {c}^{\prime } $
= −8.332), unlike the type effect (F(1,34) = 5.626, p = .023, η2 = .142; effect b = .106). This suggested that SES effects on English picture naming performance were fully mediated by caregiver input quality. A significant Sobel test (Z = −2.025, p = .04) supported this conclusion.
Figure 2. Mediation model for English picture naming.
3.4. Research question 4: Is there an association between SES, single-language measures of Maltese and English word types in caregiver input, and children’s lexical skills in each language?
To better understand the role of caregiver input quality in influencing children’s English lexical profiles, type counts were differentiated into Maltese and English words. Figure 3 shows the mean numbers of Maltese and English types employed by caregivers in each SES group, with Maltese type counts differentiated into two sub-counts, native Maltese words and established English borrowings coded as Maltese words. On average, the combined Maltese type count was significantly higher than the number of native Maltese words at every SES level (low: t(9) = 6.128, p < .001; medium: t(16) = 13.072, p < .001; high: t(10) = 13.377, p < .001). This showed that established English borrowings increased the Maltese type count significantly, so they were excluded from subsequent analyses to avoid confounding effects on children’s English lexical performance. Thus, the following results only consider the native Maltese words in the Maltese type counts.
Figure 3. Mean counts of English and Maltese types (established English borrowings and native Maltese words) in caregiver input in relation to SES level (errors bars represent 95% confidence intervals).
At each SES level, mean native Maltese type counts were significantly higher than English word types (for low SES, t(9) = 11.450, p < .001; for medium SES, t(16) = 11.382, p < .001; for high SES, t(10) = 5.599, p < .001), tying in with the predominantly Maltese-speaking home environments reported by caregivers.
Multivariate regression analysis revealed that SES significantly predicted caregivers’ use of different native Maltese words, with low SES caregivers using on average 27.655 fewer types than those having a high SES background (p = .026, η2 = .188). English types were unaffected. In turn, native Maltese type counts significantly predicted English receptive judgement (F(1,36) = 7.226, p = .011, η2 = .167) and English picture naming (F(1,36) = 5.325, p = .027, η2 = .129), while English type counts only predicted children’s English picture naming performance (F(1,36) = 5.730, p = .022, η2 = .137). Of the full range of words used by caregivers, therefore, native Maltese words were relevant to children’s English receptive judgement and picture naming performance. English types only affected English picture naming scores. Native Maltese type counts were also associated positively with Maltese picture naming performance (F(1,36) = 4.923, p = .033, η2 = .120), while English types exerted a negative effect (F(1,36) = 5.106, p = .030, η2 = .124).
As aggregate type counts were found to mediate children’s English naming skills, we sought to determine whether native Maltese type counts on their own also mediated SES effects. When SES and native Maltese types were considered together in relation to children’s naming performance, SES retained its significant effect (F(2,34) = 5.349, p = .010, η2 = .239), while the type count did not (p = .326). This indicated that on its own, the range of native Maltese words in caregiver input did not mediate SES effects on English naming skills, despite the fact that each variable separately had a significant influence on English naming performance.
4. Discussion
The scope of the present study was to investigate SES and primary caregiver input effects on the picture naming and receptive picture name judgement skills of three-year-old sequential bilingual children raised in Maltese-dominant families. The study was situated within a context of stable and nationwide bilingualism in two majority languages occurring across all SES levels. SES and input effects on children’s Maltese and English lexical performance could thus be pulled apart. Linguistic environments in which input variability is maximised and its sources are separable have been proposed as optimal for informing bilingual acquisition theory (Paradis & Grüter, Reference Paradis, Grüter, Grüter and Paradis2014). Yet, investigations of the bilingual lexical abilities of children exposed to the same language pair and having varying socioeconomic backgrounds are scarce, as are studies examining the effects of absolute measures of bilingual input quality. A consideration of these environmental factors within a single study is, to our knowledge, unprecedented.
4.1. SES effects on children’s Maltese and English lexical skills
In the current study, higher SES children performed significantly better on English lexical tasks but not on the Maltese ones. The finding of language-specific SES effects in bilingual children is not new. In bilingual contexts, a higher SES is usually associated with an increased likelihood of children receiving more, good-quality L2 input (e.g., Thordardottir, Reference Thordardottir2017), which in turn explains their increased propensity for L2 learning (e.g., Buac et al., Reference Buac, Gross and Kaushanskaya2014). Simultaneously, L1 lexical abilities may be minimally affected by SES. For example, Leseman (Reference Leseman2000) reported no SES effects on the L1 receptive and expressive vocabulary skills of Turkish-Dutch 3- and 4-year-olds. Minimal responsiveness of Spanish L1 lexical abilities to SES has also been reported (e.g., Buac et al., Reference Buac, Gross and Kaushanskaya2014). The current findings confirm that the susceptibility of L2 lexical skills to SES differences tends to occur regardless of the language pair being received in input, its majority-minority or majority-majority status and broader sociolinguistic influences related to the presence of bilingualism at a community or nationwide level.
In accordance with findings from other bilingual contexts, Maltese children’s L2 English lexical learning flourished in affluent and well-educated family backgrounds. In contrast, SES level did not predict Maltese receptive judgement and picture naming, suggesting that these skills were generally stable and not as responsive to SES, possibly by virtue of Maltese as a dominant language. Moreover, children’s receptive judgement skills in both languages were significantly better than their naming skills, at all SES levels. This is relatively unsurprising as expressive vocabulary tasks may warrant more fine-tuned language processing mechanisms when compared to receptive tasks and are usually more demanding (Gibson et al., Reference Gibson, Oller, Jarmulowicz and Ethington2012). Scores for Maltese and English receptive skills were also very similar, contrasting with findings for picture naming, where significant differences between languages emerged in the high and medium SES groups. The uneven relationship between receptive and expressive lexical performance in Maltese and English partly coincides with findings reported for 2½-year-olds receiving exposure to Spanish and English from birth, whose receptive vocabulary skills in both languages were similar, whilst Spanish expressive vocabulary scores were significantly lower than English (Ribot & Hoff, Reference Ribot and Hoff2014). A larger receptive-expressive discrepancy was likely shown by the less frequently heard language, contrasting with better English naming performance in Maltese children, despite their input being Maltese-dominant. Although Gibson et al. (Reference Gibson, Oller, Jarmulowicz and Ethington2012) identified a robust receptive-expressive gap in the L1 Spanish of 5–7-year-olds in English immersion schooling, this was likely related to a suppression mechanism operating on Spanish L1 following the onset of English immersion. The bilingual educational context experienced by Maltese pre-schoolers, however, aims to enhance children’s proficiency in both languages (Mifsud & Vella, Reference Mifsud and Vella2020).
4.2. Effects of caregiver input on children’s bilingual lexical skills
The strong correlation between absolute measures of amount and range of words directed at children replicates previous findings for monolingual caregivers (Anderson et al., Reference Anderson, Graham, Prime, Jenkins and Madigan2021; Huttenlocher et al., Reference Huttenlocher, Waterfall, Vasilyeva, Vevea and Hedges2010) as well as bilingual ones (Gámez et al., Reference Gámez, Palermo, Perry and Galindo2023) and was thus expected. Qualitative word type counts were prioritised in subsequent analyses in view of the evidence favouring their stronger influence on child language measures (Anderson et al., Reference Anderson, Graham, Prime, Jenkins and Madigan2021), their potential to mediate SES effects, unlike input quantity (e.g., Huttenlocher et al., Reference Huttenlocher, Waterfall, Vasilyeva, Vevea and Hedges2010; Stein et al., Reference Stein, Menti and Rosemberg2021), and their relevance to preschool-aged children (Golinkoff et al., Reference Golinkoff, Hoff, Rowe, Tamis-LeMonda and Hirsh-Pasek2019; Rowe & Snow, Reference Rowe and Snow2020). In this study, caregiver aggregate type counts were sensitive to SES level and predicted children’s English receptive judgement and picture naming, but not Maltese, replicating the language-specific effect of SES on children’s lexical performance. Buac et al. (Reference Buac, Gross and Kaushanskaya2014) reported a similar outcome for Spanish-English bilingual children, with only their L2 English vocabulary skills being sensitive to linguistic input variables.
4.3. Associations between SES, caregiver input quality, and children’s bilingual lexical skills
The current study’s results indicate that children’s English and Maltese lexical skills, as measured by the receptive and naming tasks employed in this study, responded differently to the same SES and input quality variables. SES affected bilingual input quality significantly, a finding that is unprecedented to our knowledge, having been documented only for monolingual input (see e.g., Hoff, Reference Hoff2003; Rowe, Reference Rowe2012). Further, SES effects on English picture naming skills were fully mediated by lexical diversity in input. Research on monolingual children has shown that qualitative aspects of language input, such as lexical diversity (e.g., Huttenlocher et al., Reference Huttenlocher, Waterfall, Vasilyeva, Vevea and Hedges2010; Stein et al., Reference Stein, Menti and Rosemberg2021) and parental responsivity (Romeo et al., Reference Romeo, Leonard, Robinson, West, Mackey, Rowe and Gabrieli2018), may mediate the effects of SES on children’s language. However, these findings are based on a limited range of socioeconomic and cultural backgrounds (Anderson et al., Reference Anderson, Graham, Prime, Jenkins and Madigan2021). In bilingual children, although there is recognition of the fact that bilingual input is shaped by the broader social context in which it is received (Hoff, Reference Hoff2020), there is limited empirical evidence examining whether SES effects are channelled through direct measures of caregiver input quality. The present findings therefore contribute towards bridging a research gap. SES influences on children’s English receptive judgement were not mediated by the word types in adult input, suggesting that caregiver-related factors other than input quality that were not measured in this study may have had a mediatory role. For example, the number of “serve and return” interactions that parents had with their 4- to 6-year-old children were found to mediate the relationship between SES and child language proficiency by Romeo et al. (Reference Romeo, Leonard, Robinson, West, Mackey, Rowe and Gabrieli2018).
4.4. Associations between SES, Maltese, and English word types in caregiver input and children’s lexical skills in each language
When adult type counts were derived for each language separately, results were even more intriguing. Across all SES levels, native Maltese type counts used by caregivers were significantly higher than English, corresponding with caregiver estimates of predominantly Maltese input directed at their children. Yet, caregivers used English words regardless of their SES level, in line with the English lexical mixing expected in Maltese child-directed language use (Gatt & Dodd, Reference Gatt and Dodd2019). Caregivers’ Maltese input was also significantly richer in the higher SES adult–child dyads. The lack of a similar SES effect on English input was possibly influenced by near-identical usage of English types by low and middle SES caregivers. Within each SES group, standard deviations approximated or exceeded the mean English type count, suggesting substantial individual variability. Maltese lexical diversity in input was subject to less variability than English, with more uniformity within SES groups probably related to very good proficiency in Maltese being reported by the absolute majority of mothers (N = 37).
Further analysis showed English types to predict English picture naming scores and native Maltese type counts to be positively associated with Maltese picture naming performance. These results fit with evidence from Spanish-English contexts showing absolute measures of lexical diversity in each language to correspond with children’s vocabulary skills in the same language (Gámez et al., Reference Gámez, Palermo, Perry and Galindo2023; Tamis-LeMonda et al., Reference Tamis-LeMonda, Song, Luo, Kuchirko, Kahana-Kalman, Yoshikawa and Raufman2014). On the other hand, it was unexpected that the range of native Maltese words used by caregivers would predict children’s English receptive judgement and picture naming. Previous research on children’s Maltese and English word use, based on parental report, hypothesised that higher-SES Maltese-speaking families provided exposure to English that was less fragmented than in lower-SES homes (Gatt et al., Reference Gatt, Grech and Dodd2016). If this were the case in the present study, however, stretches of cohesive English language use would have probably absorbed substantial proportions of the aggregate word type counts employed by caregivers. Yet, English type counts were significantly lower than native Maltese types in each SES group, suggesting that other aspects of caregiver input quality, not measured in this study, may have been at play. For example, higher lexical diversity in Maltese input might have reflected caregivers’ enhanced vocabulary knowledge. In Buac et al.’s (Reference Buac, Gross and Kaushanskaya2014) study on Spanish-English 5- to 7-year-olds and their caregivers, adults’ L2 English vocabulary knowledge was tested and found to influence children’s L2 English vocabulary performance. In the current study, potential effects of adult vocabulary proficiency seemed to operate not only within languages but also across. Higher-quality Maltese input may have conveyed richer lexical–semantic representations, enhancing children’s lexical performance in both languages. English types in input, on the other hand, only influenced children’s English naming positively. The negative association with Maltese naming skills indicated that the degree of English language mixing in Maltese-dominant exposure supported English lexical production at the expense of Maltese. This adds to the ongoing debate on whether language mixing in bilingual input exerts positive or negative influences on children’s language development (see Ruan et al., Reference Ruan, Byers-Heinlein, Orena and Polka2023).
4.5. Limitations of the study
The results of the current study must be considered in light of various limitations, including the small sample size, which highly reduces the replicability of findings to the larger population. National statistics on the number of Maltese children who are 3;04–3;08 years old, Maltese-dominant, and who do not have any developmental difficulties that affect language abilities are not available, so an accurate power analysis could not be performed. Through the use of rough population values (an estimated 1,509 children between 3;04 and 3;08 years), a margin of error of ±15.7% was derived when a confidence level of 95% was adopted. In order to elicit a much reduced (±5%) representational margin of error, the study would have had to involve an estimated 306 child participants. Although caregiver input was sampled in the home, the caregiver–child dyads were recorded interacting at a single time point in a specific play context. The 10-minute sample duration and the fixed play materials may have introduced measurement bias that limited the representativeness of caregiver input measures obtained. Anderson et al. (Reference Anderson, Graham, Prime, Jenkins and Madigan2021) noted that longer observations demonstrated stronger effects of input quality on child language outcomes. A larger-scale study could potentially investigate the quality of adult input for longer periods of time within varied contexts. A number of environmental variables that had potential to impact child performance scores, for example, birth order, gender, number of siblings, and language input received from other sources including school, were not controlled for. Similarly, aspects of input quality other than lexical diversity, namely caregiver responsivity and vocabulary knowledge, were considered as potential influencing factors but were not measured. These features would merit consideration in future studies on Maltese caregivers’ language input. Another limitation presented in the use of self-completion questionnaires, which depend on the respondents’ ability and willingness to provide truthful information about themselves. In order to simplify data coding and allow effective use of computer functions on Microsoft Excel, a number of allowances were made when categorising the lexical data derived from adult input into the respective type and token groups. For instance, the word forms depicting a direct object pronoun and a possessive pronoun could have otherwise been realised by separate lexemes (on account of the different meanings that they provide) but, for the sake of practicality, they were grouped together during analysis. Further details and examples are shown in Table A1 (Appendix).
5. Conclusion
The current study set out to investigate the effects of SES and caregiver input on the bilingual lexical skills of three-year-olds from Maltese-dominant homes. Maltese and English lexical skills responded differently to SES, with English receptive judgement and naming increasing significantly with SES, while Maltese lexical performance was unaffected. Caregiver input quality fully mediated the influence of SES on English picture naming. Consideration of single-language word type counts in caregiver input revealed the sensitivity of Maltese input quality to SES. Children’s Maltese and English naming performance and their English receptive judgement benefited from higher-quality native Maltese input that was more lexically diverse. The present findings support the notion of a cross-language effect, wherein more Maltese types in input were associated with children’s gains in English lexical skills. English type counts, representing degree of language mixing in input, complemented Maltese input quality in mediating SES effects on English picture naming but affected Maltese picture naming negatively. These results call for judicial use of language mixing.
Research involving naturally occurring mixed-language input is rare (Kremin et al., Reference Kremin, Alves, Orena, Polka and Byers-Heinlein2022). The present study may contribute to a better understanding of the role of language mixing in caregiver input and its impact on children bilingual lexical performance. While use of English words by caregivers is justified because of the benefits to children’s English lexical skills, particularly picture naming, it was found to hold back Maltese naming performance. On the other hand, caregivers’ use of native Maltese types benefited English receptive judgement and naming, as well as Maltese naming. This points towards the upkeep of Maltese language use in Maltese-dominant households, alongside English language mixing that complements rather than substitutes Maltese input. The results of this study also suggest avenues for further research. Investigations of how caregiver language mixing influences children’s mixed language responses during picture naming are called for, as well as studies that involve English-dominant adult–child dyads.
The present findings are theoretically important as they draw on two majority languages making up an under-researched language pair, in a broader context where bilingualism and language contact are the rule rather than the exception. Therefore, they have value in broadening our understanding of bilingual lexical acquisition and the extent to which it generalises or differs across language pairs and contexts of use. From a methodological point of view, it is innovative in its use of absolute measures of bilingual input quality. The positive relationship between the quality of child-directed language input and the Maltese and English lexical abilities of the Maltese-dominant three-year-old participants has implications for optimising child-directed input in homes and preschool classrooms. Language mixing in children’s input is inevitable and natural when language contact is the norm. Moreover, language mediation, or translanguaging, is a recommended classroom practice (Mifsud & Vella, Reference Mifsud, Vella and Schwartz2018). Yet, the value of high-quality Maltese child-directed input should not be under-estimated. When it includes equivalents for embedded English words, it might help boost the bilingual lexical skills of Maltese-dominant children.
Data availability statement
The Maltese and English versions of the Receptive Picture Name Judgement Task and Picture Naming Task (picture stimuli, score sheets and administration procedures) can be accessed through the following links: https://www.um.edu.mt/library/oar/handle/123456789/103057, https://www.um.edu.mt/library/oar/handle/123456789/103058, https://www.um.edu.mt/library/oar/handle/123456789/103059, https://www.um.edu.mt/library/oar/handle/123456789/103060.
The information letter used for participant recruitment and the Maltese and English questionnaires as well as the instructions and analysis code required to reproduce the analyses, including the manner by which the language samples were transcribed, how types and tokens were coded, and the computer functions used to facilitate coding can be accessed using the following link: OAR@UM: The effects of primary caregiver language input on the lexical skills of three-year-old children from different socioeconomic backgrounds.
Acknowledgements
The authors would like to express their gratitude to the editor, action editor, and anonymous reviewers for their constructive and insightful comments on earlier versions of the manuscript. They acknowledge the expert assistance of Prof. Liberato Camilleri in the statistical analysis of the data and of Dr Kenneth Grima in translating the abstract to Maltese. Their appreciation also goes to the Directorate for Research, Lifelong Learning and Employability, Malta, and to the heads of school who approved of the participation of state schools. Finally, they are very grateful to the participating children and caregivers, without whom this research study would not have been possible.
Competing interests
The authors declare none.
Appendix
Table A1. Word forms (tokens) realising the same lexeme (type)
Open access
