Cohesive devices as an indicator of L2 students' writing fluency

Investigating links between temporal features of the writing process (e.g., bursts and pauses during writing) and the linguistic features found in written products would help us better understand intersections between the writing process and product. However, research on this topic is rare. This article illustrates a method to examine associations between the writing process and linguistic features and provides an illustrative study that examines whether features of L2 writing fluency can be predicted by cohesive devices produced in argumentative essays. Seventy-five L2 undergraduate students at a U.S. university wrote two independent essays spaced about five months apart on two different prompts. Students' writing fluency was calculated using keystroke logging techniques and was analyzed using a multi-dimensional model that condensed processing into four process components (production, process variance, revision, and pause behavior). The cohesion features in students' essays were analyzed using natural language processing (NLP) approaches that measured three types of cohesive devices (reference, conjunction, lexical cohesion). Results from linear mixed effects models showed that writing fluency components were significantly predicted by a set of cohesive devices found in L2 writers' essays. The findings indicated links between certain cohesive devices and specific temporal features in L2 writers' text production (e.g., more unattended demonstratives related to more productivity). The proposed method affords opportunities for future research investigating links between the writing process and product using keystroke logging and NLP.