Model of the Classification of English Vowels by Spanish Speakers

Date
1999-10
DOI
Authors
Husaine, Fatima
Negishi, Michiro
Cohen, Michael A.
Guenther, Frank
Version
OA Version
Citation
Abstract
A number of models of single language vowel classification based on formant representations have been proposed. We propose a new model that explicitly predicts vowel perception by second language (L2) learners based on the phonological map of their native language (Ll). The model represents the vowels using polar coordinates in the F l-F2 formant space. Boundaries bisect the angles made by two adjacent category centroids. An L2 vowel is classified with the closest Ll vowel with a probability based on the angular difference of the L2 vowel and the Ll vowel boundary. The polar coordinate model is compared with other vowel classification models, such as the quadratic discriminant analysis method used by Hillenbrand and Gay vert [J. Speech Hear. Research, 36, 694-700, 1993] and the logistic regression analysis method adopted by Nearey [J. Phonetics, 18, 347-373, 1990]. All models were trained on Spanish vowel data and tested on English vowels. The results were compared with behavioral data obtained by Flege [Q. J. Exp. Psych., 43 A(3), 701-731 (1991)] for Spanish monolingual speakers identifying English vowels. The polar coordinate model outperformed the other models in matching its predictions most closely with the behavioral data.
Description
License
Copyright 1999 Boston University. Permission to copy without fee all or part of this material is granted provided that: 1. The copies are not made or distributed for direct commercial advantage; 2. the report title, author, document number, and release date appear, and notice is given that copying is by permission of BOSTON UNIVERSITY TRUSTEES. To copy otherwise, or to republish, requires a fee and / or special permission.