Talker and Acoustic Variability in Learning to Produce Nonnative Sounds: Evidence from Articulatory Training

Abstract

Compared to low‐variability training, high‐variability training leads to better learning outcomes and supports generalization of learning. However, it is unclear whether the learning advantage is driven by multiple talkers or by enhanced acoustic variability across target sounds. The current study addressed this issue in nonnative production learning. Spanish speakers were trained to produce the French /e/–/ɛ/ vowel contrast. The stimuli were recorded by five native French talkers for the multiple‐talker (MT) group or by one talker for the single‐talker (ST) group, but acoustic dispersion of the vowels and context were matched between the two groups. Both training paradigms improved production accuracy, with slightly greater improvement in the ST group. However, only MT training enhanced the compactness of vowel categories and generalized to the production of sounds elicited by an unfamiliar speaker. This suggests that talker variability supports the establishment of abstract phonemic categories in production.

The Effects of Talker Variability and Frequency of Exposure on the Acquisition of Spoken Word Knowledge

“…Results of word stress accuracy showed a general pattern supporting a stronger effect of talker variability for both stress placement and duration measures compared to frequency of exposure. This finding adds to the value of high-variability input for improving L2 pronunciation (Thomson, 2018), revealing that acoustic variability helps enhance pronunciation accuracy not only at the level of segments (e.g., Kartushina & Martin, 2019) but also at the level of individual words. Specifically, regarding stress placement accuracy, the E3 + TV condition produced significantly larger gains than the E3 condition at the immediate posttest.…”

“…There were no beneficial effects of a multiple-talker condition (three talkers) over a single-talker condition after completion of two 1-hour perception training sessions over two days. In contrast, Kartushina and Martin (2019) found that Spanish speakers with no experience with French improved production accuracy of the French mid-open and mid-close front unrounded vowels to a greater degree when they listened to the target sounds produced by five talkers than by a single talker. Wiener et al (2020) confirmed the superiority of high-variability training (four talkers) over low-variability training (single talker) for beginner-level L1 English learners studying L1 Mandarin tones after they had received explicit instruction and perception training sessions over four consecutive days.…”

“…However, recent studies looking in greater depth into the effects of talker variability suggest mixed findings regarding the extent to which high-variability training brings about larger learning gains in production accuracy compared to low-variability training (Brosseau-Lapré et al, 2013; Kartushina & Martin, 2019; Wiener et al, 2020). Brosseau-Lapré et al (2013) investigated whether English speakers with limited knowledge of French improve in their production of the French unrounded and rounded mid-vowels.…”

“…Despite the potential benefits of talker variability for learning different aspects of word knowledge, little is known about how it affects learning the pronunciation of new words. Instead, researchers in this area have generally focused on pronunciation at a segmental level (e.g., vowel contrast /e/ vs. /ɛ/ in Kartushina & Martin, 2019). Previous studies of L2 word learning and acoustic variability have also tended to rely on measures of word form learning and form-meaning mapping accuracy (using picture-to-L2-recall test) without measuring the quality of the resulting spoken forms (e.g., pronunciation accuracy).…”

Metacognition in second language speech perception and production

“…Finally, in order to assess participants' accuracy in the production of novel French vowels, similar to previous studies (Franken et al, 2017;Kartushina et al, 2015Kartushina et al, , 2016Kartushina & Martin, 2019), we computed, in a customised Matlab script, for each vowel token and subject, the Mahalanobis distance (or distance score, DS) between this token and the target acoustic space, de ned as the 20 vowel tokens used for exposure in the familiarization phase. Mahalanobis distance (Mahalanobis, 1936) is a unitless, scale-invariant measure of distance, in terms of standard deviation, from a given point to a distribution, which, by default, takes into account non-circular shape of vowel categories.…”