Search Abstracts | Symposia | Slide Sessions | Poster Sessions
Pitch memory, but not pitch aptitude, predicts implicit statistical learning of non-native tone contrasts by older adults
Poster Session B, Friday, October 25, 10:00 - 11:30 am, Great Hall 3 and 4
Yin-To Chui1, Susu Lai1, Zhen Qin1; 1The Hong Kong University of Science and Technology
Tone languages integrate segmental and tonal information to make meaning. Despite this complexity, as well as age-related decline in psychoacoustic abilities and dynamic pitch perception, older adults have been reported to demonstrate successful perceptual learning of L2 tones if explicit feedback is given. On the other hand, implicit statistical learning may pose additional challenges as it poses higher requirements on psychoacoustic and statistical learning abilities. Older learners have also been shown to recruit additional cognitive abilities to compensate for decline in sensory processing, such as phonological memory. The recruitment of compensatory mechanisms may be exacerbated in an implicit statistical tone learning task given the added difficulty. This study examined the implicit statistical learning of L2 tonal contrasts by older adults through distributional training (i.e., exposure to probability distributions of auditory tokens) and investigated the role of different cognitive factors in learning. 64 L1-Cantonese older adults (mean age: 62.9 years) learned to discriminate a perceptually difficult Mandarin level-falling tone contrast following a pre-test, training, post-test procedure. ABX discrimination was administered for pre-/post-tests, with tokens consisting of two pseudo-syllables (nua, fao) produced by two genders to test generalization. Training stimuli (female nua only) were synthesized by interpolating naturally produced Mandarin level and high-falling tones into six equidistant steps. Participants either heard a bimodal distribution (two-peak resembling level-falling categories) or unimodal distribution (one-peak resembling single ambiguous category). The bimodal group is expected to outperform the unimodal group in post-training discrimination (bimodal improvement and unimodal suppression in support of distributional learning). A cognitive battery was also administered that measured the learners’ pitch aptitude (just-noticeable-difference of pitch contour), pitch memory (average tone span), and working memory (average backwards digit span). Mixed-effects logistic regression models were performed on participants’ accuracy in the ABX discrimination tasks, with fixed effects of Distribution (Unimodal vs Bimodal) and Session (Pre-test vs Post-test). Results showed a main effect of Session (z=3.74, p<.001), suggesting that both groups improved through training. However, the interaction between Distribution and Session was not significant (z=-0.49, p=.622), suggesting no divergence between bimodal/unimodal improvement. Results for cognitive predictors showed that the three-way interaction between Distribution, Session, and Predictor was significant only for pitch memory (z=-2.08, p=.037). Post-hoc simple slopes analysis revealed that pitch memory predicted improvement in the unimodal group only. Specifically, higher pitch memory resulted in worse post-training improvement (i.e. better suppression from unimodal exposure). Both analyses suggest the results are driven by the performance of the unimodal group. Contrary to expectation, the unimodal group did not suppress improvement through exposure to a distribution that facilitated the formation of a single category. This was explained through large individual variability in pitch memory – only individuals high in pitch memory showed suppression. The lack of group results highlights a difficulty in implicit statistical learning of tones for older adults, where only “good learners” (high in pitch-related memory capacity) may be able to store the distributional information for subsequent tonal discrimination. This ability is perhaps more important during unimodal exposure because of the lesser acoustic salience of the tokens.
Topic Areas: Speech Perception,