‘Eating’, ‘drinken’ or both?

Dutch and English exposure effects on the structure of young bilinguals’ receptive vocabulary networks.

Abstract

Network analytical approaches have been applied to the study of young bilinguals’ word learning strategies by a growing body of research in various settings (for a review, see Wojcik, 2018). They have investigated the effect of one language on the structure of the other’s network or compared both networks, trying to identify similarities. These studies have identified features that influence bilinguals’ acquisition process: frequency, phonological features (Bilson et al., 2015), the presence of translation equivalents and cognates (Bosch & Ramon-Casas, 2014), and familiarity (Wojcik, 2018), among others.
The present study was an exploratory attempt to capture the structure of the receptive vocabulary networks of young bilingual children, by combining both of their languages. Network analysis can model the connections or links of word-nodes based on selected lexical features. The links in the present social networks were determined by how many of the participating children knew two words at the same time. This knowledge was previously tested with the English (Dunn & Dunn, 1997) and Dutch (Schlichting, 2005) versions of the Peabody Picture Vocabulary Test (PPVT) with children with early exposure to both languages (Aalberse et al. [MIND-team], 2021). Exposure to Dutch and English at children's homes and daycare centers for each individual child was assessed through a parental and daycare questionnaire respectively (MIND-team, 2021). Social network analysis, using the ORA-lite software (Altman & Reminga, 2018), was used to create a network for all the children (N=297; MAge= 34.5 months, SDAge= 6.5 months) and two networks of two subgroups based on exposure (Group A: higher English than Dutch exposure, and Group B: higher Dutch than English exposure). Average exposure scores were used as a threshold and cut-off point to determine ‘high’ and ‘low’ exposure groups. The Leiden algorithm was applied to the networks (Traag et al., 2019). The algorithm is designed to identify groups, sets, and clusters within dense networks with many nodes and links.
The comparison of these three networks revealed the role of exposure in acquisition and hints at the strategies employed by bilingual children in word learning. The total group of children, without exposure factored in, showed clear clustering along the language divide, with Dutch words being more central in the network. To account for exposure, the total group was divided into two subgroups, one with higher exposure to Dutch (N=118; MAge= 34.84 months) and one with higher exposure to English (N=100, MAge= 34.23 months). Whereas the subgroup with higher Dutch exposure mirrored the clustering trends of the total group of children, the subgroup with higher English exposure showed higher cross-language clustering, with both English and Dutch words being in the same Leiden cluster more consistently in the network. In the group with higher Dutch exposure, the three node-sets detected, through the Leiden algorithm, were 1) central Dutch words, 2) peripheral Dutch words and 3) peripheral English words. Contrastively, for the group with higher English exposure, the central node-set included words from both languages. This marks a group of words (from both languages) with increased learnability and potential links with each other.
Joint membership in frequent semantic categories (animal names, food, clothing, small household items, and body parts) and phonological features (word length and initial sound) are the most likely candidates that guided cross-language clustering. Word difficulty affected the structure of the network as a whole: words belonging to higher PPVT sets were known by fewer children than the ones in lower sets.
Overall, exposure scores were shown to have an important effect on cross-language clustering and network structure. To which relative the children spoke to and for how long they spoke to them per day (MIND-team, 2021) could determine the centrality of words in the network. Children growing up in the Netherlands have a high exposure to Dutch, as the societal language. That meant that the English group (higher English exposure) had more exposure to Dutch than the Dutch group (higher Dutch exposure) had to English. For the English group, that led to a more balanced exposure to both languages, which most likely led to higher cross-language clustering for learned words.
This study offers a new methodology for studying lexical acquisition, with the aim of capturing the interconnected nature of bilinguals’ languages. The results from the balanced exposure group are in accordance with emergentist accounts of language acquisition (Claussenius-Kalman et al., 2021), suggesting that ‘one word [might utilize] the conceptual packaging invoked by a word in the other language' (Bilson et al., 2015). There were semantic and phonological features that linked words across languages in learning. No strong effect of cross-linguistic learning was found for translation equivalents and cognates, because these items were rare between the English and Dutch versions of the PPVT. Nonetheless, the contributions of the present study are an added step to the investigation of the factors that might drive connectivity in lexical acquisition.