From pauses to processes: Automating the classification of L2 writing processes from keystroke logs

Although cognitive models of writing have long emphasized the importance of planning, formulation, and revision, little is known about how adult L2 writers distribute these subprocesses during real-time composition. This study addresses that gap by investigating the temporal and spatial characteristics of writing behaviors in a timed argumentative task. Sixteen Spanish university students participated in a 35-minute English writing task, recorded with Inputlog. Writing events were automatically classified using a VBA-based system that integrates pause duration, pause location, and event context to identify underlying subprocesses. The analysis examined how these subprocesses evolved over time, how they related to linguistic pause locations, and how pausing behavior differed across subprocesses when measured as pauses per minute, mean pause duration, and relative pause frequency. Results indicated that subprocess distribution remained relatively stable throughout the session, but clear associations emerged between pause location and subprocess type, for instance, within-word and before-word pauses were linked to formulation, while sentence- and paragraph-initial pauses were associated with planning. Differences in pause duration and frequency also reflected distinct cognitive demands, with planning marked by longer pauses and revision by shorter, more frequent ones. The results support the classification system and shed light on cognitive effort in L2 writing, with potential applications for feedback and task design.