USING NATURAL LANGUAGE PROCESSING TO ADVANCE SECOND LANGUAGE WRITING ASSESSMENT: EVIDENCE FROM CORPUS-BASED RESEARCH

Abstract

This paper explores how Natural Language Processing (NLP) can support the assessment of second language (L2) writing, particularly in relation to textual coherence and cohesion. The study centers on the DECOR (Detect, Explain, and Rewrite) framework as a tool for identifying discourse-level weaknesses and generating revisions that improve the organization and connectedness of learner writing. Data for the analysis are drawn from the EF-Cambridge Open Language Database (EFCAMDAT), a large-scale learner corpus containing more than one million English texts produced across different CEFR proficiency levels. Because the corpus includes extensive learner metadata and draft histories, it offers a valuable basis for examining patterns of writing development over time. Methodologically, the research adopts a mixed approach that combines automated analysis and revision through DECOR, corpus-informed feature extraction, and human evaluation of writing quality. The effectiveness of NLP-based intervention is measured through comparisons between original and revised drafts, with particular attention to gains in coherence and lexical cohesion. The findings reveal common discourse-related difficulties among L2 writers and show that NLP-driven feedback can make a meaningful contribution to writing improvement. By connecting computational analysis with established human rating practices, the study underscores the educational value of AI-assisted writing assessment and promotes scalable, evidence-based, and learner-oriented methods for multilingual writing instruction.