Collocation processing in translation and writing between Chinese and English: a corpus-based and keylogging analysis

This study investigates how collocations are produced and processed in bilingual text production by native Chinese speakers, focusing on L1 Chinese and L2 English. Collocations, as multiword units, facilitate fluent native-like language and reflect cognitive effort during production (Feng, 2020; Ferraresi & Bernardini, 2023). The research aims to examine (1) holistic processing of collocations versus non-collocations, (2) L1–L2 differences in collocation processing, (3) task effects (writing vs. translation), and (4) the role of typing fluency. Eighteen Chinese MA students completed writing and translation tasks in both languages. Collocations were operationalized by syntactic dependencies and corpus-based association measures (t-score, MI, logDice), while pauses, inter-keystroke intervals over 2000 ms, were categorized into five patterns (_X_, PX_, _XP, PXP, XPX) , originally proposed by Dahlmann and Adolphs (2007) and further refined by Gilquin (2024a), which reflect their placement relative to the collocation. Mixed-effects logistic regression analyses show that collocations are more likely than non-collocations to be produced without internal pauses, supporting holistic processing. L1 collocations involve fewer pauses than L2, and writing elicits more pauses than translation. Fluent typing is associated with reduced within-collocation pauses. Descriptive analyses highlight the dominance of the _X_ pattern and the potential of logDice as a corpus-based collocation metric. By integrating corpus- and keylogging-based approaches, the study sheds light on real-time collocation processing and demonstrates the methodological value of keystroke logging for research in phraseology and second language acquisition.