Advances in the Assessment of Young Bilinguals: Comments on Floccia et al

This study of vocabulary size in bilingually-developing 2-year-olds addresses two important and timely questions. The first is whether the rate of early vocabulary growth in bilingual children differs depending on which two languages the children are acquiring. The second is whether an instrument can be developed to assess vocabulary growth and identify language impairment in bilingual children. Findings that address the first question should reveal something about how the language acquisition process works. Findings that address the second question should have widespread significance for clinical practice with children who are exposed to more than one language. Thus, the research reported in this monograph is important for its potential to shed light on the nature of the language process, and it is important and timely for its potential to address a clinical needs that becomes more pressing as the number of children in dual and multilingual environments grows.

No commentary can do justice to the enormity of this undertaking. Researchers from six different universities in the United Kingdom collaborated to collect data on the language development and language environment of 430 bilingual 2-year-olds. Apart from the project’s goals of identifying effects of typological distance and developing norms for the assessment of bilinguals, this project stands as the largest study of simultaneous bilinguals in the literature. (There was some variability in age of exposure to English, but at age 2 years the range of variability is necessarily small.)

Project summary

The research reported in this monograph takes advantage of the natural experiment created by the many different language backgrounds of families with young children in the UK. Using online recruiting through a website aimed a families of bilingual toddlers, the researchers identified 372 2-year-old children who were exposed to English and one of 13 other languages: Bengali, Cantonese, Dutch, French, German, Greek, Hindi/Urdu, Italian, Mandarin, Polish, Portuguese, Spanish, and Welsh, and 58 children who were exposed to English and an additional language not on this list of non-English languages. The significance of this list of non-English languages is that an assessment instrument of the same form as the English assessment instrument was available for each. The instruments were modified versions of the MacArthur-Bates inventories in which caregivers report, from a list of words provided, the words that their children understand and the words that their children produce. This data collection was done online, and it provided the outcome measures of receptive and expressive vocabulary.

The candidate predictor variables were of two sorts: (1) properties of the children and their language experience, and (2) the linguistic distance between English and the other language each child was hearing. Properties of the children and their language experience were measured via questionnaires completed by the caregiver on line and in a telephone interview with a researcher. These measures included information about how much the child heard each language in child-directed and overheard speech, who used each language with the child, the child’s exposure to language in day care, and more. The measures of linguistic distance included indices of phonological overlap between words in each language that had the same meaning, similarity of word order typology, and similarity in terms of morphological complexity.

Study 1 focused on testing the effect of linguistic similarity between a child’s two languages, after the effects of other identified sources of variance in children’s vocabulary size were statistically removed. The analyses identified three variables that met the authors’ criteria of having a statistically significant relation to the outcome and accounting for at least .02 percent of the variance in outcome: gender, the proportion of English in child-directed speech, and the proportion of English in overheard speech. The effect of linguistic distance was tested separately for each measure of linguistic distance by asking if model fit was improved when the measure of linguistic distance was added to the base model, which included child gender and the two measures of English exposure. The findings were that phonological overlap improved model fit in accounting for expressive vocabulary and similarity in word order typology and morphological complexity improved model fit in accounting for receptive vocabulary. In both cases, similarity had a positive influence, and in both cases the benefit of similarity between English and the children’s other languages were observed only for the other languages, not for English.

Study 2 focused on developing the clinical instrument, the UK Bilingual Toddlers Assessment Tool (UKBTAT). This instrument consists of the 100 word, short form Oxford Communicative Development Inventory (Hamilton, Plunkett, & Schaefer, 2000) with gender- and language exposure-adjusted norms for interpreting bilingual children’s scores on those inventories. The 430 2-year-olds who were assessed in English provided data for the model of English vocabulary skill that yielded the adjusted norms. The 372 2-year-olds who were also assessed in their non-English language provided data for assessment of these non-English languages (excluding Spanish for lack of permission). The assessment tool uses the norming data to predict for any individual child what his or her vocabulary score should be when the model is provided data on the child’s gender and English language exposure. The model yields predicted expressive and receptive vocabulary scores in English and in the non-English language. The difference between a child’s predicted and observed scores provides the basis for calculating the child’s percentile. The unique feature of this percentile score—and the purpose of this part of the project—is that it is the child’s percentile among children of the same gender and with the same amount of language exposure. The amount of exposure children have to a language is the single strongest predictor of their rate of development, and adjusting norms for exposure has been argued to be a necessity in the assessment of bilinguals (Gathercole, Thomas, Roberts, Hughes, & Hughes, 2013).

Study 3 tested the model, recalculated with 372 children as the database, against the 58 children not included in the database. This test of the predictive ability of model found that the model’s predicted scores correlated with the children’s actual scores with r’s of .60 for comprehension and .59 for production.

Contributions to understanding cross-linguistic influences in bilingual development

The monograph’s findings of effects of linguistic distance are new findings in an area where little is known. There are previous studies of speech sound perception in bilingual infants exposed to languages that differ in the acoustic boundaries that define phonemic categories (Werker & Byers-Heinlein, 2008). Some, but not all, of these studies find that bilingual infants take a little longer to solidify their phonological categories when their two languages divide up the acoustic space differently (see Hoff, 2015). There are also previous studies of cross-linguistic influences of grammatical properties of each language on second language acquisition in childhood. These studies find that children are helped in learning structural properties of a second language that are shared with their first language. Children whose first language marks tense and agreement on the verb do better at English verb morphology when acquiring English as a second language, compared to children whose first language has no verb morphology (Blom, Paradis, & Duncan, 2012; Dulay & Burt, 1974; Paradis, 2011). Children whose first language has no articles make article omissions when acquiring English (Zdorenko & Paradis, 2011).

In both of these previous lines of research on crosslinguistic influences in bilingual development, the domain of the predictor and the outcome were the same. In contrast, in Floccia et al., the search is for influences of phonological and grammatical similarity between the two languages on vocabulary growth. The question is what makes word learning easier or harder in two languages being acquired at the same time. It appears that similarity between the two languages being acquired, both in word order and in morphological complexity, make acquiring a receptive vocabulary easier—with the direction of influence going from the majority language to the minority language. The explanations offered for this finding refer to the benefit to the child of being able to use similar processes to segment the speech stream where word order is similar and of being able to apply the same learning approach to morphology where the morphological complexity is similar. The idea that the nature of the language a child is acquiring shapes the child’s approach to the task has support in the literature (e.g., Estes & Bowen, 2013; Hay, Graf Estes, Wang, & Saffran, 2015). Filling in the details of how such a process operates in bilingual development is clearly an area for further research. The evidence in this monograph also indicates that phonological overlap between languages makes acquiring an expressive vocabulary easier. Floccia et al. interpret this as consistent with the advantage in lexical access for cognates that has been observed in older bilingual children and adults, and, in particular, it is consistent with the argument that bilinguals are better at accessing words with cognates in their other language because those words are acquired earlier than other words (Costa, Pannunzi, Deco, & Pickering, 2016).

The question is why should cognates with phonological overlap be acquired earlier. Core and colleagues argued that part of the word learning task is representing the phonological form of the word (Core, Hoff, Rumiche, & Señor, 2013). Where there is phonological overlap, the work of creating that phonological form supports creating a lexical entry in both languages. Other findings make a similar argument. Schelletter (2002) studied one German-English bilingual child and found that the lag between the child’s beginning to use a word in one language and beginning to use the equivalent of that word in the other language was related to the similarity of the phonological forms of the words. To illustrate, the lag between the appearance of tree and Baum was greater than the lag between the appearance of apple and Apfel in his productive vocabularies. Among 18-month-old Spanish-Catalan bilinguals, Bosch and Ramon-Casas (2014) found evidence that the phonological similarity between the two languages gave these bilingual children a boost in vocabulary acquisition. When words with identical forms were counted for each language, the bilingual children had significantly larger total vocabularies than monolingual children. When the duplicate forms were subtracted from the total, the bilingual and monolingual children had similarity vocabulary sizes. Together, these studies suggest that simultaneous bilingual children get new words at reduced cost when they are phonetically similar to the equivalent word in the other language; when the two words have identical forms the children get two for price of one. It makes sense that the benefit would apply more to expressive than receptive vocabulary, if the sort of phonological representation required for comprehension is less difficult to achieve than the sort of representation required for production (Bates, 1993).

In sum, this project gives us the beginning of an answer to question of whether bilingual development is different depending on the particular languages begin acquired. The answer is yes, it is. The research reported in this monograph is a start at filling in the details of how and why.

Contributions to the assessment of bilingual children

The development of the UKBTAT is potentially a major advance for the clinical assessment of bilingual children. The UKBTAT address two problems that have historically made assessment of bilingual children difficult: (1) There are not norms for bilingually developing children, and the norms developed on monolingual children are not appropriate for evaluating bilingual children, and (2) When one of the languages a bilingual child is acquiring is a relatively unstudied language, instruments and norms for that language do not exist.

The problem of establishing norms for bilingually developing children is a challenge because bilingual children not only differ from monolingual children in development, they are also quite heterogeneous themselves. It has long been recognized that the inadequate database on normative bilingual development has resulted in bilingual children being both under- and over-identified as delayed in language development (Bedore & Peña, 2008; Gathercole, 2013). The UKBTAT addresses the first problem by providing norms adjusted for features of children’s learning experience with demonstrated effects on performance on that instrument. The UKBTAT addresses the second problem, to a degree, because the experience-adjusted norms allow evaluating a bilingual child’s language learning ability based on what the child has learned in only one of his or her languages.

The UKBTAT does have limitations. One is that the UKBTAT is applicable only to 2-year-olds. Floccia et al. make a compelling argument for the value of a screening instrument at age 2, but one can easily imagine that older children might also need evaluation. A second limitation is that there are not tests for every non-English language, nor is it clear how clinicians are to interpret two scores when a score for each language is available. The methodological innovations in the research reported in this monograph point to ways to address these limitations.

Methodological innovation

Several methodological innovations which supported this enormous endeavor also provide a means for extending the work. One innovation is the collection of caregivers’ reports of children’s vocabulary and background information on line and over the phone. The efficiency of this procedure makes more feasible the development of new instruments, including instruments to assess older children. Another innovation and efficiency is the creation of short (i.e., 30-word) instruments across languages and doing so by translating the English word list. Although Floccia et al. found ceiling effects with the 30-word English inventory they used, it might be possible to develop inventories larger than 30 items but still sufficiently small to be easy to develop in multiple languages.

The reasonableness of assessing bilingual children using short lists of cross-language equivalents in their two languages is supported by findings of substantial commonalities across languages and across rates of language growth in the content of children’s vocabularies. Rescorla and colleagues have used a different vocabulary checklist, the Language Development Survey (LDS, Rescorla, 1989), to assess vocabulary knowledge in young monolingual children in English, Greek, Polish, European Portuguese, and Italian. The first words produced by children from these different language backgrounds overlapped substantially (Rescorla, Constants, Bialecka-Pikul, Stępień-Nycz, & Ochal, 2017; Rescorla, Nyame & Dias, 2016). The first words of typically developing children and late talkers also are highly similar (Rescorla, Alley, & Christine 2001). This high level of overlap in words in the early vocabularies of children acquiring different languages supports the proposal to select a core set of words that could be translated into multiple languages and used in paired assessment instruments to identify bilingual children at risk of ongoing language delay.

There are arguments against such an approach. It does not take account of differences among languages and among language pairs in the rate of typical development, which have been demonstrated to exist (Bleses et al., 2008; O’Toole et al., 2016). Generating word lists for new languages ignores differences among cultures in the things children talk about (Hoff, 2013a), which is why this is not an approach used in developing the MacArthur-Bates instruments in multiple languages. But the goal in this case would be different from the goal in developing MacArthur-Bates instruments. The goal is a less fine-grained and sensitive measure of developmental level; it is a screening device to identify risk. Furthermore, as Floccia et al. argued, cultural differences in what children talk about are likely to be diminished when both languages are used in the same environment, as is certainly more the case for bilingual children than for two children speaking different languages in different countries.

Having available instruments for assessing both languages that a bilingual child is acquiring yields two advantages: (1) There would be an instrument available to assess children whose English (or other majority language) skills cannot be measured. Although the UKBTAT adjusts norms for the amount of the language exposure, there are likely to be children whose English (or other majority language) experience is minimal, making assessment based on the majority language unreliable. Furthermore, for children in households where little majority language is spoken, there may not be a family member who can provide information on the child’s majority language vocabulary comprehension or use. (2) The scores on two parallel instruments could be considered together to assess conceptual, and total vocabularies as well, should clinicians wish to do so.

There are limitations to the online procedure, which would remain. Not every caregiver has access to the internet or the facility to use such a platform. This, in conjunction with the online recruitment of participants, likely contributed to unequal representation of children from lower socioeconomic strata in Floccia et al.’s sample. Similarly, online collection of caregiver reports may not work for all children in need of clinical assessment.

Contributions to the study of early bilingual development

Apart from the investigation of effects of linguistic distance, and apart from the development of an assessment instrument, the research reported in this monograph stands as the largest study of 2-year-old bilinguals in the literature. It is worth considering what this study adds to our knowledge of the nature of dual language proficiency at this age in children who have been exposed to two languages from birth (or close to that) and what this study tells us about sources of influence on early bilingual development.

The overall description of the dual language proficiencies of these 430 2-year-olds living in the UK and exposed to English and another language is that they have smaller vocabularies than monolingual children (based on comparison to the UK norms for the MacArthur-Bates instrument) while also having larger vocabularies in English (the majority language) than in their heritage (minority) language. This description is entirely consistent with findings from studies of other immigrant groups including children of Spanish-speaking families in the U.S. (Hoff, Core, Place, Rumiche, Señor, & Parra, 2012; Hoff & Ribot, 2016), children of Russian-Finnish bilingual families in Finland (Silvén, Voeten, Kouvo, & Lundén, 2014). There are many reasons for this outcome, but it is clear from these data that early exposure to two languages does not necessarily start children on two monolingual-like growth trajectories.

Other descriptive findings in Floccia et al. were that the bilingual children experienced balanced English and non-English input on average but they were English dominant in their vocabulary knowledge, on average. Additionally, their receptive vocabularies were larger than their expressive vocabularies. Similarly, studies of children in Spanish-speaking immigrant families in the U.S. that balanced input at home leads to English dominance in the children and that children with balanced dual language skills comes from minority language-dominant homes (Hoff, 2017, Hoff, Core, Place, Rumiche, Señor, & Parra, 2012; Hoff & Ribot, 2017) and that the gap between comprehension and production can be even greater in bilinguals than in monolinguals (Gibson, Oller, Jarmulowicz, & Ethington, 2012; Gibson, Peña, & Bedore, 2012; Ribot & Hoff, 2014; Ribot, Hoff, & Burridge, 2017).

The strongest significant positive predictors of English vocabulary in these two-year-old bilingual children were gender (with the advantage to girls), relative amount of English in child-directed speech, and relative amount of English in overheard speech. The gender effect is consistent with previous findings in monolingual and bilingual development (e.g., Bornstein, Hahn, & Haynes, 2004; Galsworthy, Dionne, Dale, & Plomin, 2000; Huttenlocher, Haight, Bryk, Seltzer, & Lyons, 1991). The effect of relative amount of exposure in child-directed speech is probably the most robust in the literature on early bilingual development (Hoff, 2017; Unsworth, 2016). It is fully consistent with the argument that language acquisition is a process of learning the regularities in the language system from their manifestation in input. One can, and scholars do, argue about the nature of the representation of those regularities and the nature of the learning process. But the fact that language acquisition depends on language exposure remains a stubborn fact in both the literature on monolingual development and bilingual development (Hart & Risley, 1995; Hoff, 2006, 2013b, 2017).

The finding of a positive effect of the proportion of input from native English speakers on children’s English productive vocabulary is also consistent with previous findings (Place & Hoff, 2011; 2016). In Floccia et al.’s model, native speaker input accounted for only 1% of the variance in English vocabulary. This finding is also consistent with the evidence from Place and Hoff (2011, 2016) that the percent of English input from native speakers is strongly correlated with the amount of English input, and a unique effect of native input is not consistently observed across measures.

The finding of a positive effect on children’s English of the degree to which the parents used both English and the non-English language is consistent with findings of effects of exposure to Spanish and English in the U.S. The findings in both the U.S. studies (Place & Hoff, 2011; 2016) and in this monograph were interpreted as reflecting the likelihood that language mixing by heritage language speakers serves to increase exposure to English. It is important to point out that measure of language mixing in this monograph was parents’ answer to the question of whether they use only one or use both languages with their child. It was not a measure of the intrasentential code-switching that has been argued to interfere with acoustic cues to word boundaries that might make word segmentation and thus word learning more difficult (Byers-Heinlein, 2013).

The finding of a significant effect of exposure to English in overhead speech is not wholly consistent with the extant literature, although that literature is not itself wholly consistent. Some studies have demonstrated that children at age 2 years can learn from overheard speech (Ahktar, 2005). Some studies have found no evidence of learning from overheard speech (Weisleder & Fernald, 2013; Schneidman & Goldin-Meadow, 2012). At least one study has found that the ability to learn from overheard speech depends on the child’s age, with 24 months and a certain amount of vocabulary knowledge being something of a dividing line between those who cannot and those who can (Ma, Golinkoff, Houston, & Hirsh-Pasek, 2011). In addition to child age and vocabulary level, it has been suggested that children in some cultures may be better than children in other cultures at monitoring and learning from the conversations of others (Akhtar, 2005). Also, there are likely differences among overheard speech. Overhearing speech among others about the here-and-now is potentially more useful than overhearing one side of a telephone conversation. In this research, estimates of amount of overheard speech in each language were reported by caregivers, so it is not clear what sort of overheard speech was counted.

Although null effects are always to be interpreted with caution, given the sample size of this study it is worth noting some effects that they did not find. Some are more interesting than others. They did not find effects family income or parental education, but as the authors point out, the range on these variables was limited. They did not find an influence of siblings after language exposure effects are accounted for. This is not surprising, as the authors point out, because the sibling effect in the literature is only of older school aged siblings and is mediated by siblings influence on English language use in the home (Bridges & Hoff, 2014).

This monograph revealed no effect of children’s attendance at English-language child care on their English language vocabulary. This is a potentially meaningful finding, and it is also consistent with a reading of the literature that concludes that while preschool programs can boost children’s language, the benefit of preschool depends on the quality of the preschool and many show much smaller effects than the best known programs (Burchinal, Zaslow, & Tarullo, 2016; Duncan & Magnuson, 2013). This dependence on features of what happens inside the preschool is also true for the benefit of preschool for bilingual children, and although it might seem obvious that sending a child from a minority language home to a majority language preschool should improve the child’s majority language skills, that is not always the case (Bowers & Vasilyeva, 2011; Lauro, Hoff, & McAdams, 2017). Peers also matter (Hoff, 2006). If the majority of the children in the class are similarly from minority language homes, perhaps it is not surprising that majority language skills do not measurably benefit.

There was no effect of the proportion of the non-English language input from native speakers on vocabulary in that language. This too is not surprising, given that the mean proportion of non-English from native speakers was 95%. As was found in studies of Spanish-English bilingual toddlers in the U.S., exposure to English comes from a mix of native and nonnative speakers but exposure to the heritage language is almost entirely from native speakers.

There was no effect of language the status of the non-English language. The bilingual development of English and Welsh was not different from the bilingual development of English and any of the other non-English languages in the sample, although Welsh enjoys a social status and support from the larger community that the other non-English languages do not. It has been argued that heritage language acquisition is vulnerable because of its lack of community support, and that bilingual development is more successful when both languages enjoy a similar level of social prestige, as do French and English in Canada (Smithson, Paradis, & Nicoladis, 2014). Here, no such effect was observed, but 2-year-olds may be less sensitive to prestige differences than older children.

Conclusion

The research reported in this monograph contributes important new descriptive data on early bilingual development to a field that is still small in numbers of studies and often small in sample sizes. More interestingly, this research raises provocative questions about how the task of bilingual development may be different depending on which two languages are being acquired. Most centrally, the work reported here has produced a new and better instrument for assessing the language learning capacity of children whose language exposure and resultant language knowledge is distributed across two languages.

The most significant innovation in the development of that instrument is the establishment of norms for single language vocabulary size that are corrected for the amount of exposure to that language the child has experienced. It has been said that an SRCD Monograph should end an existing field of research or begin a new one. This monograph has the potential to begin a new field in which instruments are developed to assess individual children based on empirically derived models of what children in their circumstances can be expected to know. The method is applied here to the assessment of language ability in bilingual children; it has the potential for broader application to other domains in which we must infer the nature of a child’s internal capacities based on their reflection in the child’s achievement.