As is the case in the current study, vocabulary knowledge is regularly assessed using specifically designed tests. It is possible to assess both the understanding of words (receptive vocabulary) through the selection of an appropriate illustration from a group of 3–4 pictures and the ability to verbally provide a vocabulary verbally item by naming a picture (expressive vocabulary). These methods of assessing vocabulary do not rely on contextual cues, and although one may argue that this does not represent real-world vocabulary understanding and use, decontextualised language is an important characteristic of academic language (Rowe 2019).

For the purpose of this study, bilingual children are defined as ‘those who use two (or more) languages in their everyday life’ (Armon-Lotem & De Jong 2015:5), and this does not imply equal proficiency in the languages.

The assessment of vocabulary in both languages spoken by bilinguals is important because research has frequently shown that bilinguals have a smaller vocabulary in each of their languages than monolinguals speaking the same language (Burridge & Stebbins 2016; Hoff et al. 2012; Thordardottir et al. 2006). However, because of the distributed nature of lexical knowledge, assessing bilinguals’ vocabulary in only one language is likely to underestimate their proficiency (Gross, Busc & Kaushanskaya 2014). As Bialystok et al. (2010:7) point out: ‘the world is being constructed through two telescopes for bilingual children, and their two vocabularies provide the lenses’. It is important to capture this dual vocabulary knowledge. However, the challenge in assessing bilingual children is that some of their vocabulary overlaps across languages and some may be language-specific (Bedore et al. 2005). Thus, there may be concepts for which the bilingual knows a corresponding word only in language A and others only in language B (Gross et al. 2014). This is as a result of context-specific learning, resulting in exposure to certain lexical items in only one language. This is referred to as ‘singlet vocabulary knowledge’ (Gross et al. 2014:575). Other areas of vocabulary may be experienced in both languages and result in ‘doublet vocabulary knowledge’ (Gross et al. 2014:575). Bilingual children may have a larger proportion of singlets than doublets in their vocabularies (Mancilla-Martinez & Vagh 2013; Sheng, Lu & Kan 2011).

Consequently, Pearson, Fernandez and Oller (1993) developed a procedure known as conceptual scoring to capture bilingual vocabulary more accurately. This technique considers the total number of concepts for which a child knows a word, irrespective of the language in which the word is known. For example, an isiZulu-English child may know the word for ‘shoe’ in both English and isiZulu [isicathulo], but only know unogwaja [rabbit] in isiZulu and ‘diamond’ [idayimane] in English. This would result in a conceptual vocabulary score of 3. Conceptual scoring has been found by several researchers to boost the measured vocabulary of bilingual children to the level of their monolingual peers and to be useful over a range of bilingual language experiences (Anaya, Peña & Bedore 2018; Bedore et al. 2005; Marchman, Fernald & Hurtado 2010). In addition, Marchman and Martinez-Sussman (2002) established that bilinguals’ conceptual vocabulary score was positively correlated with their spontaneous language production.

However, conceptual scoring may be affected by test-specific factors (i.e. language and modality) and child-specific factors (i.e. acquisition history and socio-economic status) (Gross et al. 2014). The language for which the vocabulary test was designed (i.e. the child’s first or second language) or the administration procedure may influence the conceptual score. Thus, translated tests assessing the same vocabulary in both languages may yield different results from tests where the vocabulary in the two language tests is different and based on the types of words to which the bilingual child may be exposed in each of his or her languages. In addition, if the administration of the test allows responses in either language, conceptual scores may be different from those obtained when only single-language responses are required. With respect to modality, conceptual scores may vary depending on whether receptive or expressive vocabulary is considered. Children may understand many more words than they can express, and thus the receptive conceptual score is likely to be higher than the expressive score. Therefore, it is important to assess both receptive and expressive vocabularies.

Bilingualism develops in different ways and a distinction is often made between simultaneous acquisition, when both languages are learnt from infancy, and sequential acquisition, when exposure to a second language occurs after infancy, usually at the start of preschool or first grade (Kohnert 2010). In either case, the duration and quality of exposure to each of the languages will contribute to proficiency and possibly also to the conceptual vocabulary.

The design of the study was quantitative, descriptive, cross-sectional and comparative in nature. Numerical data comprising the vocabulary scores obtained by the participants in both English and their home languages (isiZulu or Afrikaans) were analysed using statistical procedures (Schiavetti & Metz 2006). The study was descriptive in that no deliberate experimentation or manipulation of variables occurred, and cross-sectional in that the data were collected at a single point in time (Hegde 2004). The study also included between- and within-group comparative components in that the vocabulary scores of the monolingual and bilingual participants were compared with each other and the scores of the bilingual learners were compared across their languages.

The study was conducted at two primary schools in Johannesburg, Gauteng province, where the language of teaching and learning is English from Grade 1. The schools were purposefully selected to ensure that there were sufficient numbers of monolingual English speakers as well as bilingual IsiZulu-English (site 1) and bilingual Afrikaans-English (site 2) learners. Both schools were former model C schools in the northern suburbs of Johannesburg. Learners attending the site 1 school are from the surrounding suburbs where their parents either live or work. The socio-economic status of the learners is varied. The majority of teachers at the school are first language English speakers. A variety of home languages were reported for the learners at the school. These included, but were not limited to, isiZulu, English, Sesotho, Sepedi, SiSwati, Xitsonga, Setswana, isiXhosa, French and Portuguese. Thus, in any given classroom, there is significant linguistic diversity and support for learning in the home language may be limited. Learners at site 2 were mostly from middle- and high-income families. Similar to the situation in site 1 school, the majority of teachers at this school were first language English speakers, but the number of first languages reported for the learners was less diverse and included only Afrikaans, English, isiZulu, isiXhosa and Portuguese.

A non-probability, purposive sampling strategy was employed by approaching the specific schools and inviting their participation.

Participants were required to be healthy with no obvious cognitive, physical, hearing or visual impairments that might impact language development. At site 1, 8-year-old learners in Grade 2 were selected to ensure that they had sufficient exposure to English in a school environment at the time of assessment. At site 2, 7-year-old children in Grade 1 were selected, as their home and preschool exposure to both English and Afrikaans was considered to be sufficient to cope with the assessment. The bilingual learners were required to have IsiZulu (site 1) and Afrikaans (site 2) as their primary home languages, respectively. Confirmation of isiZulu or Afrikaans as a first language and English as a second language was established through the parent consent forms and school records. Children from either gender were eligible to participate in the study. Two monolingual English groups were included to serve as a comparison with the bilingual groups at each school. These participants were selected according to the same criteria (no obvious disabilities and ages 8 or 7) as the respective bilingual groups, but were required to speak only English at home.

Table 1 summarises the composition of the samples.

As shown in Table 1, the monolingual and bilingual groups at each site were matched for age and there was an equal distribution of male and female children at both sites 31 female and 29 male children in site 1 and 30 female and 30 male children in site 2.

The instruments used to assess the vocabulary of the participants in this study were:

These instruments were selected based on previous research into vocabulary in bilingual individuals showing sensitivity to subtle vocabulary differences (Allman 2005; Pearson et al. 1993). The ROWPVT-4 and EOWPVT-4 are considered to be reliable and consistent measures of vocabulary (Allman 2005). These vocabulary tests were also chosen above others as they have been translated into several languages, suggesting their suitability for adaptation into other languages. For example, Allman (2005) used the English and Spanish standardised versions of the EOWPVT to compare English and Spanish monolinguals and bilinguals. Dionysios et al. (2010) adapted both tests into Greek for the school-aged Greek population and found the tests to be sufficiently sensitive. Chiang and Rvachew (2007) used the English EOWPVT and an adapted French version to assess vocabulary of bilingual children in Canada. The ROWPVT and EOWPVT have not been used extensively in research in South Africa and so the validity of their use in this context can be commented on in the present study.

The ROWPVT-4 assesses an individual’s ability to identify objects, actions and concepts based on a spoken word when given a choice of four illustrations (Martin & Brownell 2011a), for example, show me aquatic. This task targets the understanding of words and is referred to as receptive vocabulary. Cueing, prompting and picture clarifications are not allowed (Martin & Brownell 2011a). Some adjustments were made to test items in the ROWPVT-4 in the present study to make them more culturally appropriate to the South African context. Linguistically equivalent words were used to ensure that the integrity of the test was not compromised. Table 2 shows which English words were changed to better suit the South African context.

The EOWPVT-4 assesses the ability to name objects, actions and concepts when presented with illustrations and is referred to as expressive vocabulary (Martin & Brownell 2011b), for example, what is this? or what is he doing? Only prompts specified in the scoring manual for each item were used to elicit a response from the participants, for example, what is this? for a singular object and what is one word for all of these for a group of objects such as fruit. Additional cueing was used if the participant was not attending to the correct part of an illustration, for example, labelling ‘sky’ for the illustration of ‘cloud’ as a result of inattention to the arrow pointing to the cloud. Additional cueing was not used if an item was labelled incorrectly, for example, labelling a cloud as ‘rain’. Additional cueing was given if the participant responded with a sentence or a phrase, where a single word was required. Misarticulations were not scored as incorrect. Where applicable, a variety of responses were acceptable and listed on the response form, for example, rug/carpet/mat. One adjustment was made to the list of acceptable responses, that is, mielie for ‘corn’ to make it more culturally appropriate to the South African context.

In both the tests, repetition of the target word is allowed and neutral feedback is given after each response. The participants were not given the correct answer for incorrect responses in order to avoid influencing future performance, specifically by the bilingual participants. Both the ROWPVT-4 and the EOWPVT-4 are individually administered tests and were developed to be used on ages 2–80+ years (Martin & Brownell 2011a, 2011b). Thus, these tests were considered suitable to assess the expressive and receptive vocabulary of 7- and 8-year-olds for the purpose of this study. Both tests have basal and ceilings that are guided by age and the number of correct or incorrect responses. For this study, basals and ceilings were not used as the raw data collected in this study would not be compared with the normative data provided in the test manuals. Participants completed all items up to number 110. This self-imposed ceiling was chosen based on the average raw scores achieved by 7–8-year-olds according to the ROWPVT-4 and EOWPVT-4 formal data.

The English ROWPVT-4 and EOWPVT-4 were adapted into isiZulu and Afrikaans. The method of testing one language at a time using two separate tests, that is, an English test and an isiZulu/Afrikaans test, was supported by Anaya et al. (2018), who found that this method resulted in higher scores and a more accurate conceptual vocabulary score. This was in contrast to using one test and prompting for responses in either one or both languages at the same time. Anaya et al. (2018) found that scores are negatively affected by the demands of code-switching between the two languages. Despite the limitations of translated tests, the English tests were translated into isiZulu and Afrikaans in this study, mainly because of a paucity of research on the vocabulary to which isiZulu and Afrikaans children are exposed, which would allow for different tests to be constructed in these languages. In addition, the vocabulary assessed in the English ROWPVT-4 and EOWPVT-4 was considered to be basic and fairly universal in terms of the concepts assessed.

Nonetheless, adaptation and administration of the tests were carried out with the aid of isiZulu and Afrikaans-English bilingual translators. The English words were given to one translator to translate independently into isiZulu or Afrikaans while another translator was asked to back-translate the isiZulu or Afrikaans words into English to assess the validity of the translations. This process was repeated twice and acceptable words and variations were deliberated on by all parties. While translating and back-translating may provide well-translated materials, cultural and linguistic equivalence must also be considered (Peña 2007). Many of the English words in the ROWPVT-4 and EOWPVT-4 did not have direct isiZulu translations and so were not included in the isiZulu assessments. This resulted in the isiZulu ROWPVT-4 and EOWPVT-4 consisting of 88 and 102 items, respectively. Care was taken to ensure that the words were linguistically equivalent, for example, ‘slumber’ was not adapted to ‘sleep’ in the isiZulu version. In contrast, it was possible to translate all the test items into Afrikaans, resulting in the Afrikaans versions consisting of 110 items.

Each participant was tested on the school grounds at minimal disruption to the academic schedule. The adapted isiZulu/Afrikaans tests were administered by trained first language isiZulu/Afrikaans speakers who were familiar with children, vocabulary testing and the research process. The testing was supervised by the researchers. Scoring and prompting were the same as for the English EOWPVT-4 and ROWPVT-4. The only notable prompting differences were for the expressive testing, where children required inconsistent reminders to provide the isiZulu/Afrikaans name, rather than a slang word where a word is adapted from another language, for example, itrain for ‘train’ where isitimela is the correct word in isiZulu. The Afrikaans-English bilinguals tended to provide the English word and had to be prompted for the Afrikaans word.

Testing lasted approximately 30 minutes for both the expressive and receptive portions of the assessments in each language. Monolingual learners were only assessed in English. The isiZulu/Afrikaans bilinguals completed half the receptive and expressive vocabulary test in English first, and then half in isiZulu/Afrikaans and vice versa, so as to counterbalance the effects of familiarity.

External validity in this study was addressed by having an adequate sample size of 30 monolingual and 30 bilingual learners at each site. A small sample of the population at each site (n = 10) was scored by both an examiner and a researcher to establish inter-rater reliability. There was 100% agreement in scoring in these instances. A small sample size (n = 10) was also tested twice with 100% agreement between the two tests, resulting in a Pearson’s correlation coefficient of 1, confirming test–retest reliability. All the isiZulu and Afrikaans scores were reviewed by a second L1 isiZulu/Afrikaans speaker and there was 100% agreement between scores on all 60 assessments at each site. Raw data entries were checked independently to ensure accurate data capturing.

Participants’ scores for each item on each test were entered onto spreadsheets, containing an assigned number for each participant, gender and first language. The raw scores for each monolingual and bilingual group on each test were used to calculate a standard mean, standard deviation and range of scores. It is important to note that this is a standard mean based on peer scores and not on the standard norms as published with the EOWPVT-4 and ROWPVT-4 manuals. Mean scores were converted to percentages for ease of comparison. Conceptual receptive and expressive vocabulary scores were calculated for each participant using the method described in Table 3.

Within the group, comparisons were performed using paired sample t-tests to compare the receptive and expressive isiZulu/Afrikaans and English scores in the bilingual groups and the English receptive and expressive scores in the monolingual groups. The Pearson correlation coefficient was performed to determine correlations between receptive and expressive measures in each group of monolingual and bilingual participants at each site.

Independent sample t-tests were used to compare the monolingual and bilingual groups at each site on the English ROWPVT-4 and EOWPVT-4 scores as well as the monolingual English scores and the bilingual conceptual scores at each site.

All of the between- and within-group statistical comparisons were deemed significant if the probability (p) of rejecting the null hypothesis was < 5% (p < 0.05).

Approval and ethical clearance to conduct the study were obtained from the Human Research Ethics Committee (Non-medical) of the University of the Witwatersrand (clearance number: H14/06/29 [site 1] and H16/07/39 [site 2]). The proposed study was presented to and approved by the Gauteng Department of Education. The principals at the identified schools gave permission for participation and the respective teachers at each school were given information regarding the study and were asked to assist with the identification of suitable participants. An information letter and consent form were sent home with selected learners who met the criteria for participating in the study. Learners participated only if a parent or guardian had given written consent. The researchers explained the study in a child-friendly manner to the learners who had returned completed consent forms and if the learners gave assent, they were included in the study. Participation was entirely voluntary. Participants’ anonymity was protected by assigning numbers and not using their names in the data capturing.