Fig. 4.

Accurate decoding of L2 proficiency from EEG data. (A) A multilinear principal component analysis (MPCA) was performed on the TRF weights corresponding to speech descriptors at all linguistic levels of interest. The first MPCA component was retained for the TRFs corresponding to Env, Phn, Pt, and Sem. The combination of these four features was predictive of L2 proficiency (r = 0.68), with significant effects for all features that were not due to group differences in age or attention. (B) A support-vector regression analysis shows that EEG data accurately predicted the L2 proficiency level at the individual subject level (r = 0.83, MSE = 1.14). (C) Classification accuracy for L1 versus L2 and L1 versus C-level L2. The red dotted lines indicate the baseline classification levels, which were calculated as the 95^th percentile of a distribution of classification accuracies derived after randomly shuffling the output class labels (N = 100). (For interpretation of the references to color in the text, the reader is referred to the web version of this article.)