Individual differences in working memory impact the trajectory of non-native speech category learning

Abstract

The ability to learn novel speech category contrasts is an important skill in second language learning. There is substantial individual variability in the ability to learn perceptual categories. Prior research demonstrates that higher working memory capacity is associated with better initial category acquisition, typically assessed within a single session of learning. There is mixed evidence for the role of working memory in speech category learning and the underlying mechanisms are not well understood. To better understand the role of working memory in speech category learning beyond initial acquisition, we trained participants on non-native Mandarin tone speech categories across three sessions, separated by one and two months, respectively. Examining all participants together, we found that working memory was associated with better speech category learning from initial acquisition to learning months later. However, when considering only participants who performed at above-chance levels in the task, we found that working memory was positively related to performance only in initial sessions and not in later learning. Working memory was positively associated with generalization of category knowledge to novel talkers and was unrelated to maintenance of category knowledge across sessions. Using longitudinal drift diffusion mixed models, we found that higher working memory was associated with more efficient evidence accumulation rates throughout learning and more cautious responding in later learning sessions. These results indicate that better working memory is not a guarantee of enhanced speech category learning, but it may reflect quicker and more efficient initial learning, which may be less effortful for a learner. Similarly, lower working memory does not doom a learner to poor performance but may instead be linked to higher risk of task disengagement and slower initial learning.

Combining pupillometry and drift-diffusion models reveals auditory category learning dynamics

“…The remaining iterations were thinned in intervals of 5 to reduce autocorrelation. Consistent with prior research, we focus on the DDM parameters associated with correct responses only because gradual improvements in correct categorical decisions characterizes learning (Roark et al, 2021(Roark et al, , 2022. Posterior means are reported as point estimates and 95% pointwise credible intervals were used to assess uncertainty.…”