Dissociating age of acquisition from proficiency: On the use of linear mixed-effects modeling of ERP data in native speakers and bilinguals

Newman, A. J. , Tremblay, A. & Nichols, E. S.

Dalhousie University, Halifax, NS, Canada

While later age of acquisition (AoA) of a second language (L2) leads to generally lower proficiency, variability also increases. This creates confounds for analyses that treats L2 learners as a homogeneous group. Because AoA and proficiency tend to be highly collinear, separating these factors is challenging. Approaches that treat continuous measures like AoA or proficiency as categorical variables may further reduce statistical power. It is also critical to characterize the effects of proficiency in native learners to understand whether effects in L2 learners reflect similar variation in language processing or qualitatively different effects associated with L2 acquisition.
We report two event-related potential (ERP) studies analyzed using linear mixed-effects modeling. In Experiment 1, we presented lexical-semantic violations to L1 and late L2 learners of English (AoA at least 18 years). Proficiency was measured with a standardized test (TOAL-3) in all subjects. N400 amplitude, but not latency, was influenced by proficiency in both native and late learners across the scalp, though in native learners the effect was strongest at left anterior electrode sites. The collinear effects of learner status and proficiency were addressed by residualizing one of these variables against the other. Once proficiency was accounted for, L1/L2 status did not add significant predictive power to the model. The second experiment employed morphological priming using regularly- and irregularly-inflected French verbs. Participants were L1 and L2 learners of French (AoA at least 5 years). Proficiency was measured using a conjugation test. N400 amplitude increased with proficiency for all words. Morphological priming effects (N400 amplitude reduction for regular infinitive forms preceded by inflected forms) were found to be largest for low proficiency individuals, becoming smaller as proficiency increased.
These studies emphasize that proficiency affects ERPs associated with lexical processing, and suggest ways in which the collinearity of proficiency and AoA may be disentangled.