AI-ASSISTED ERROR TAXONOMY IN MEDICAL ESP WRITING: A CORPUS-BASED STUDY
AI-ASSISTED ERROR TAXONOMY IN MEDICAL ESP WRITING: A CORPUS-BASED STUDY

Summary/Abstract: This study investigates the use of AI-assisted annotation for corpus-based error analysis in medical English for Specific Purposes (ESP) writing. The corpus comprised 318 essays written by second-year medical students (85,981 words). Errors were initially identified and classified using a large language model (LLM) according to a predefined taxonomy and subsequently reviewed through human validation. A total of 4,472 errors were identified, corresponding to an average of 5.52 errors per 100 words. Grammar was the most frequent error category, followed by register/formality and vocabulary. Recurring patterns included subject–verb agreement errors, tense inconsistency, lexical imprecision, and inappropriate academic register. In addition to general language errors, ESP-specific issues involving medical terminology, discipline-specific collocations, and medical academic style were found in approximately 18% of the essays. A small but statistically significant negative correlation was observed between essay length and normalized error density (r = –0.18, p = 0.0014). The findings indicate that medical ESP learners face both general linguistic challenges and discipline-specific writing demands. They also suggest that AI-assisted annotation, when combined with human review, offers a practical and scalable approach to corpus-based error analysis in specialized educational contexts.

• Details
• Contents