Model of the Classification of Second Language Phonemes

The paper presents a vowel classification model that explicitly predicts vowel identification by second language (L2) learners based on the phonological map of their first language (L1). The model represents vowels in formant space by polar coordinates with the mid-range of the formants as the origin. The model assumes that category boundaries bisect the angles made by two adjacent category centroids. It further assumes that the probability of an L2 vowel being classified as a L1 vowel depends on the angular difference between the L2 vowel formants and the L1 vowel boundaries. The model was trained on Spanish vowels and tested on English vowels, and it outperformed other models in predicting the behavioral data.