Assessing accuracy and variability based on empirical thresholds in French oral vowels produced by Korean speakers

Previous research on assessing L2 speech has largely focused on evaluating accuracy using the mean values of acoustic features, while comparatively less attention has been paid to articulatory variability, which reflects the consistency and consolidation of phonological categories in L2 production. This study investigates the articulatory accuracy and variability of ten French oral vowels (/i, e, ɛ, a, u, o, ɔ, y, ø, œ/) produced by ten Korean learners compared to those produced by six native French speakers. Each vowel was produced 12 times per speaker under a 2 (/b/- vs. /p/-initial CV word) × 2 (one vs. two CV words in the carrier sentence) × 3 repetitions design, resulting in 1,920 tokens. Five acoustic features were analyzed for accuracy (vowel duration, F1, F2, F3, vowel space area), and five for variability (VarcoV, formant trajectory length, formant dispersion, F1/F2 and F2/F3 ratios). Thresholds were empirically defined based on just noticeable differences from prior studies or, when unavailable, based on statistical dispersion of native speakers’ data. Results showed that in the accuracy dimension, significant threshold exceedances were observed for /e/ (F1), /ɔ/ and /ø/ (vowel duration), and /y/ (F2, F3). All other vowels (/i, ɛ, a, u, o, œ/) remained within threshold limits despite statistical differences in some features. In the variability dimension, exceedances occurred for /ɔ/ (VarcoV), /ø/ (formant trajectory length), /y/ (formant dispersion), /i/, /e/, /ɛ/, /ø/ (F1/F2 ratio), and /e/, /u/, /ɔ/ (F2/F3 ratio). Accordingly, /e/, /ɔ/, /y/, and /ø/ exceeded thresholds in both accuracy and variability, confirming that temporal, spatial, and proportional dimensions operate independently in explaining articulatory variability. These findings demonstrate that the proposed threshold system effectively differentiates perceptually relevant articulatory differences that are not captured by statistical significance alone, thereby providing a multidimensional basis for quantitative evaluation of L2 vowel production.