ABSTRACT
This article reports on an investigation into the bilingual development of literacy skills in Portuguese heritage language speakers in Switzerland in which we sought to put Cummins’ linguistic interdependence hypothesis to the test. Three predictions were derived from this hypothesis: (1) Literacy skill levels in one of the bilinguals’ languages can best be predicted from skill levels in both languages at an earlier point in time; (2) Crosslinguistic effects originating from the language in which literacy skills are explicitly taught are stronger than those originating from the language in which they are not explicitly taught; (3) Crosslinguistic effects are stronger between more closely related language pairs. These three predictions were evaluated using longitudinal reading and writing data from 233 French–Portuguese and German–Portuguese bilingual children. Results show that prediction (1) is borne out but predictions (2) and (3) are not. The discussion centres on the vast array of possible results that are compatible with the interdependence hypothesis in correlational studies, particularly if it is combined with post hoc thresholds. Our conclusion is that current forms of the interdependence hypothesis do not offer a fruitful framework for further investigation.
Disclosure statement
No potential conflict of interest was reported by the authors.
Notes on contributors
Raphael Berthele is Professor in multilingualism at the University of Fribourg. He directs the MA programmes in multilingualism studies and in foreign language didactics and he co-founded the Fribourg Institute of Multilingualism in 2008. His research interests cover different areas from cognitive to social aspects of multilingualism. During the last years, he has been focusing on the empirical investigation of receptive multilingualism and on convergence phenomena in the semantic and syntactic patterns in linguistic reference to space in multilinguals.
Jan Vanhove is a senior assistant at the Department of Multilingualism in Fribourg, Switzerland. He has done research on receptive multilingualism, crosslinguistic influence between closely related languages and lexical richness. He blogs semi-regularly about research design and statistics in applied linguistics and multilingualism research at http://janhove.github.io.
ORCID
Raphael Berthele http://orcid.org/0000-0003-1403-4116
Notes
1 Ardasheva, Tretter, and Kinny (Citation2012) a priori operationalised the lower L2 threshold relative to a state-administered standardised test. They did not, however, test for the presence of a non-linear developmental slope with an inflection at this threshold.
2 Both Brunner and Austin (Citation2009) and Westfall and Yarkoni (Citation2016) suggest that researchers incorporate measurement errors in their statistical models, for example, using structural equation modeling (SEM). Such models have been used in interdependence studies and related research (e.g. Gebauer, Zaunbauer, and Möller Citation2013; Schoonen et al. Citation2011; Verhoeven Citation1994) but typically require multiple indictors for each construct, which the present project does not have. Additionally, even when multiple indictors per construct are available, the causal relationships in SEMs may still be overstated: in longitudinal interdependence studies, the measurement errors are likely to be correlated both across languages and across different points in time inasmuch as the same or similar tasks are used. As Reddy (Citation1992) demonstrates, ignoring such correlations may also cause the causal relationships in the SEM to be overstated. Unfortunately, estimating all these additional parameters would quickly exhaust the data available. In sum, we do not see any quick solutions to establish whether cross-linguistic longitudinal effects are due to interdependence (or another structural factor) or simply a by-product of measurement error. Resolving this issue is a clear desideratum for future studies.