ABSTRACT
Theories of multilingual language learning often assume that previous language learning experiences in at least two languages provide advantages for additional language learning. Other research emphasises the importance of general cognitive abilities as predictors of language learning. We test whether a set of predictions based on a multilingual advantage theory or on a cognitive abilities-related theory fit the data better. Whereas many studies in both traditions focus on (young) adults, our study focuses on multilingual language learning in children. We discuss evidence from the primary school context in German-speaking Switzerland. The 115 pupils from 8 different schools were in 7th grade at the first measurement time (T1) and in 8th grade one year later (T2). They are predominantly German-speaking, and they learn first French and then English as foreign languages at school. German and French skills were tested at T1, and English skills one year later at T2. In addition, verbal and visuo-spatial working memory and general intelligence were tested at T1. Two structural equation models are fitted to the data, representing two different theoretical assumptions. Both models are acceptable but adding interdependence of the three languages does not substantially increase the fit to the data.
Acknowledgement
We would like to thank Carina Steiner, Jan Vanhove, and Amelia Lambelet for their help during data collection and analysis. We thank the following MA Students in the Fribourg Department of Multilingualism for their help with the preparation of the data collection and for their valuable fieldwork: Thomas Aeppli, Alessandra Dedei, Kinga Dobrowolska, Laura Hodel, Rachel Howkins, Patricia Isler, Alexandra Jaszkowski, Bente Lowin Kropf, Nina Müller, Heike Reimann, Pauline Robert-Charrue, Fabio Soares.
Disclosure statement
No potential conflict of interest was reported by the authors.
Notes
1 Common European Framework of Reference for Languages.
2 Numbers are given for week 29 of the school year. As data collections at T1 and T2 were held within a time frame of three weeks, numbers may vary slightly.
3 This is calculated using (startlengh + total correct)/trials per length.
4 Proficiency in the three languages can theoretically also be modelled as three latent variables, with the different subscores of the tests as manifest variables. We decided to simplify the analysis here and use one summary score per language as manifest variables. The alternative analysis was also carried out and it yielded the same overall result (see supplementary material for details).
5 MLR in lavaan refers to a maximum likelihood estimation with robust standard errors and a scaled test statistic. Missing values are deleted listwise in the structural equation model and pairwise in the descriptive plots in . We did not impute the data since missing values were not missing at random.