Computational perception of information foci produced by Chinese English learners and American English speakers

This study used computational perception, via SVM and Random Forest models, to examine phonetic features used by American English speakers (AE) and Chinese second language learners of English (CE1 with low proficiency and CE2 with high proficiency) in realizing different information foci. For all participant groups, the machine learning models achieved above chance level accuracy. Coda duration and the duration of the rising contour were two phonetic features that ranked top across three participant groups in terms of their importance to the models. The SVM models trained with the AE data classified different foci by CE1 and CE2 with above chance level accuracy, but English proficiency had little effect on the classification results.