Enhancing Indonesian EFL Students’ Academic Writing Proficiency through NLP-Based Writing Assistants: A Mixed Methods Study
Abstract
Artificial Intelligence (AI) integration into language learning has attracted increasing attention, particularly for enhancing the academic writing of English as a Foreign Language (EFL) students. However, limited research has examined how NLPbased writing tools support both surface-level accuracy and higher-order writing development within pedagogically guided environments, especially in the Indonesian EFL context. This study aims to investigate the effectiveness of NLP-based writing assistants and explore students’ learning experiences. Utilizing an explanatory sequential mixedmethods design, the study involved 100 university students enrolled in an academic writing course. The quantitative phase employed a one-group pretest-posttest quasi-experimental design, while the qualitative phase involved semi-structured interviews. Writing performance was assessed using an analytic rubric covering grammar, cohesion, coherence, and argument structure. Data were analyzed using paired-sample t-tests and thematic analysis. Results revealed a statistically significant improvement in students’ writing performance (p < 0.001), with a large effect size (Cohen’s d = 1.87). Improvements were observed across all writing components, particularly in coherence and cohesion, indicating enhanced organization of ideas alongside linguistic accuracy. Qualitative findings showed increased linguistic awareness, greater engagement in iterative revision, and improved writing confidence. However, students reported difficulty interpreting higher-order feedback on cohesion and argumentation. While findings indicate the potential of NLP-based writing assistants, causal interpretations remain limited due to the absence of a control group. This study highlights the importance of integrating automated feedback with pedagogical guidance to optimize learning outcomes and recommends further research using more rigorous and longitudinal designs.