Acoustic distances, Pillai scores and LDA classification scores as metrics of L2 comprehensibility and nativelikeness

Evaluating L2 pronunciation via acoustic measures is problematic. In the literature, this is sometimes accomplished via fluency metrics (speech rate, average length of IPUs, etc), which however do not capture pronunciation accuracy. Other studies use VOT measurements (only possible if L1 and L2 differ in this respect) or comparison of vowel formants with native values. This contribution uses vowel distances in an F1-F2 space, Pillai scores and classification scores for vowel pairs as acoustic metrics of L2 vowel pronunciation. We compute these metrics on speech produced by 25 learners of L2 English and we compare them with (a) fluency metrics computed on the same productions, (b) VOT measurements, (c) impressionistic judgments provided by native speakers. The advantage of this approach is that L2 pronunciation accuracy is not judged in reference to comparable native productions, but intrinsically: it measures the extent to which phonological vowel contrasts are kept apart in L2 speakers’ realisations.