Using data from bilingual children to test models of online word comprehension: Results and future directions

• In Event: Bilingualism - early childhood

Abstract

During the first years of life, children develop the ability to understand familiar nouns (Bergelson & Swingley, 2012; 2015; Fenson et al., 2007) as they mature and are incrementally exposed to language. Most theories about young children’s vocabulary development highlight the importance of both maturation and experience, but typically view one factor as a more important driver (Bergelson, 2020). For example, current computational models of vocabulary development emphasize the role of accumulating experience with words (e.g., McMurray, 2007). However these models have only been tested using parent-report group level data from vocabulary production. It is thus unclear if those same models would also emphasize experience accumulation when tested in other ways, for example using individual-level data, or when vocabulary is measured via online comprehension. In a novel approach, we used individual-level data from bilingual infants in the looking-while-listening paradigm to test models of vocabulary development. Contrary to monolinguals, who mature and learn their language simultaneously, bilinguals accrue different levels of experience with two languages while they mature, allowing us to disentangle these factors in a way that is typically not possible for monolinguals.
Method
We used archival data from five different looking-while-listening (LWL) studies (2012–2019). Children saw two images on a screen: a target (e.g., dog), and a distractor (e.g., book), while one was labeled (e.g., “Look at the dog!”) and eye gaze was recorded.
We retained data from 155 bilinguals and monolinguals (14–48 months) – with a continuum of experience with the target language ranging from 10% to a 100% (e.g., a child could hear 20% French, 80% English and be tested in one or both languages). We compared the roles of maturation (age) and experience (% exp) as predictors of looking accuracy through four linear mixed effects models representing plausible word learning models: maturation only, experience only, additive-maturation plus experience, and accumulator-maturation times experience. The best-fitting model was the additive model in which both maturation (age) (p < .001***) and experience (p = .001**) were independent contributors of noun comprehension.
Discussion
We tested vocabulary development models using individual-level LWL data. We found the main driver of vocabulary development to be the additive effects of maturation and language experience. Furthermore, our study provides insights about conducting LWL studies with bilingual populations as we found that childrens’ performance is affected by even minimal changes in language experience, and that monolingual children tend to outperform bilingual peers when only one of their languages is considered.

Our results are in contrast with previous models that emphasize the role of experience accumulation rather than the additive effects of maturation and experience. We believe that the individual-level data from LWL may have provided nuance not previously captured by computational models. This highlights the importance of conducting more research using individual-level data and with different measures to test models of vocabulary development. As such, we are currently collecting data for a replication and extension study. These new data will allow us to 1) examine word learning in younger infants, 2) look at maturation more precisely, and 3) encompass a broader cross-linguistic comparison (English, French and Spanish).

Authors

• Andrea Sander-Montant, Concordia University
  Presenting Author

• Melanie López Pérez, Concordia University

• Krista Byers-Heinlein, Concordia University