Abstract
This study is the first to examine predictors of oral and written second language (L2) chunking development among older adults over time. Based on a two-year longitudinal design with time-series data, we set out to determine how and when individual differences in prior L2 exposure and working memory (WM) predict the frequency and association strength of adjective–noun bigrams in both the written and oral modes. We analyzed L2 chunking development among 43 older adult learners aged 61–70 who completed a two-year L2 English course in Switzerland (first language German), each of whom participated in 33 consecutive data collection waves over the instructional period. Unsurprisingly, high-frequency bigrams dominated oral speech compared to written texts, but exclusive constituent words (measured via Mutual Information [MI] scores) were not more prevalent in one mode than the other. Generalized additive modeling revealed that prior L2 exposure was only a weak predictor of chunking development, while complex WM emerged as the strongest predictor, particularly for written production, where learners with higher WM capacity produced more frequent and tightly associated bigrams. We also observed a temporally limited effect of complex WM, which supported collocational exclusivity early in training but later reversed, suggesting shifting cognitive involvement as learning progressed. This study offers timely insights into how a long-neglected demographic acquires and utilizes formulaic language and which individual learner differences contribute when to L2 chunking development.