Convolutional neural network applications to identifying differences in phoneme processing in EEG signals of native speakers and late second language learners

Language processing differs in native language speakers and language learners, and making progress encapsulating these differences in brain responses is difficult due to the complexity of the brain. In particular, late second language learners are posited to differ in learning in comparison to native or bilingual speakers, for instance in phonological knowledge. Deep Learning Neural Networks (DLNN) offers the opportunity to capture nuances of electroencephalographic (EEG) data that could be less perceivable using ERP methods. Using DLNN analyses of EEG data could be another tool sensitive for identifying and categorizing the data. The present study used a DLNN to categorize EEG data from L1 (native) and LL2 (late 2nd language learners) English speakers while they listened to English phonemes. A subset of participants from both L1 and LL2 groups was used to train the DLNN and then another different subset of participants was used to test the DLNN performance on categorizing single-trial EEG of L1 participants from EEG of LL2 participants. Overall the results showed that the trained DLNNs were reasonably accurate at categorizing novel speakers with 73% ± 8% accuracy. This may be evidence that there are general processing differences between native speakers and learners of a language.Medicine, Faculty ofAudiology and Speech Sciences, School ofGraduat