2024/06/24 by Rachel Meyer, Ratree Wayland, Kevin Tang +2
#acoustic signal processing #machine learning #neural networks #phonetics #phonology #second language acquisition
paper · doi:10.7275/scil.2210
Lenition, or the weakening of stop consonants, is a prevalent phonological feature of Spanish. This study investigates the time course of acquisition of lenition throughout a study abroad program. Unlike previous studies on lenition, this research measures the degree of lenition using posterior probabilities of the continuant and sonorant phonological features, as estimated by the deep learning model, Phonet. Results show that the posterior probability for the [continuant] feature increases for learners during the study abroad program, but the [sonorant] posterior probability does not. Taken together, these results suggest a fricative production, rather than the more native-like approximant production. Phonet's estimates of the posterior probabilities for continuant and sonorant phonological features appear to be reliable measures of lenition.