Training Child Learners On Non-Native Vowel Contrasts with Phonetic Training: The Role of Task and Variability

Abstract

Acquiring a non-native speech contrast is often difficult. High variability phonetic training (HVPT) is a well-established method used to train learners on non-native phoneme contrasts: it critically uses high variability (HV) input after earlier studies using low variability (LV) input had proved unsuccessful. HVPT has since been successfully applied in many different studies with adult participants, but much less so with children. The current study further investigates the effect of input variability on phonetic training for children, examining whether, like adults previously, they show a HV benefit. In a two-week training study, 58 Dutch 7-year-olds and 51 Dutch 11-year-olds were trained on Standard Southern British English vowel contrasts using either HV (multiple talkers) or LV (single talker) input. Both groups improved in training, however, contrary to the idea of a plasticity benefit for younger learners, only the 11-year-olds showed evidence of generalization i.e. improvement on pre/post-tests with unfamiliar talkers, with Bayes Factors showing null evidence for 7-year-olds. Critically, there was no evidence that 11-year-olds showed greater generalization following HV input, with evidence for the null in some tests. Results are discussed in terms of difficulties of the training and testing tasks and possible interactions with the benefits of input variability.

Does high variability training improve the learning of non-native phoneme contrasts over low variability training? A replication

“…Floor effects would have likely been less of a concern given our learners are adults with some English experience. Furthermore, a similar kind of phonetic training paradigm but with 7-and 11-year-olds who had very little English experience showed improvement over a set of 8 training sessions, even if pre-tests scores were at chance (Brekelmans et al, 2020). Given that 1) our learners generally had more experience of English and 2) there is no reason to suspect that phonetic training per se only works for certain pre-test accuracy, we did not include a lower cutoff for pre-test scores.…”

“…Finally, Brekelmans et al (2020) also investigate the role of high versus low talker variability in 7-year-olds and 11-year-old Dutch children learning English vowels. They found that both age groups improved as a result of eight training sessions, but 7-year-olds did not show any generalisation abilities regardless of which input variability condition they were in.…”

“…However, some of the studies discussed also looked at performance during the training stage, generally finding stronger performance for single talker than multi-talker input (c.f. Brekelmans et al, 2020;Dong et al, 2019;Giannakopoulou et al, 2017;Sinkevičiūt ė et al, 2019;Sadakata & McQueen, 2014;Perrachione et al, 2011). We saw above that some researchers have suggested that there may be a trade-off between the benefits of high variability for generalisation and the cost of processing multiple talker input during phonetic/vocabulary training (Perrachione et al, 2011;Sinkevičiūt ė et al, 2019;Sadakata & McQueen, 2014).…”

“…One difficulty in interpreting the mixed findings is that where null findings are reported, only Dong et al (2019) and Brekelmans et al (2020) have incorporated statistical methods (Bayes Factors) which allow us to evaluate the evidence for the null. Other studies with null results used frequentist p-values as their inferential statistic, which means we are not able to conclude whether those studies that fail to find evidence for a high variability benefit actually provide evidence for the nullthey may simply be underpowered to detect an effect.…”

“…Although this is not reported in the original Logan et al (1991) and Lively et al (1993) experiments, as discussed above, a single talker benefit during training has been seen in other phonetic training studies (cf. Evans and Martín-Alvarez, 2016;Evans et al, 2017;Giannakopoulou et al, 2017), including in Dong et al (2019) and Brekelmans et al (2020) which used similar blocked approaches to variability in order to remove the difficulty of trial-by-trial adaptation. The hypothesis of a single talker benefit during training is tested to shed light on (2) and to help to interpret that result.…”