Analyse de la production de voyelles anglaises par des apprenants francophones, l’acquisition du contraste /ɪ/–/iː/ à la lumière des k-NN

Abstract

Can methods of supervised learning help to evaluate the degree of acquisition of a contrast in a second language? This article addresses the issue of assessing the realizations of /ɪ/ and /iː/ by French learners of English in spontaneous conversations by considering it as a classification problem, which the resort to the k-nearest neighbours algorithm (k-NN) can solve. From a longitudinal corpus of 12 speakers recorded at the university of Paris Diderot over two years, it is shown that in spite of gaps in the numbers of occurrences of tokens for each category, and in spite of the impossibility to use cross-validation, the k-NN method is a reliable means to account for the quality of learners’ vocalic realizations.

Keywords

• French learners
• pronunciation model
• automatic detection
• k-NN.