Enhancing L2 Testing through AI-Assisted Automated Item Generation

초록

While automated item generation has gained attention across various disciplines, its application in English language teaching (ELT) remains relatively new. AI-assisted item generation has evolved from template- and rule-based models to statistical and deep learning approaches. Transformer-based systems (e.g., GPT) now produce more authentic and context-sensitive items, though issues of validity, fairness, and psychometric rigor persist. This study offers a comprehensive overview of automated item generation for language educators, drawing on a review of existing literature. It outlines major approaches, synthesizes findings from ELT-focused studies, and introduces four AI-powered tools: ChatGPT-4, genQue, Perplexity AI, and Twee, with practical guidance for classroom use. Findings indicate that ChatGPT generates items through prompt-based adaptation and iterative refinement. At the same time, Perplexity AI integrates GPT-4 with real-time search to provide source-based responses, though its mechanisms remain unclear. genQue, designed for the Korean College Scholastic Ability Test (CSAT), creates structured passages and items at varying difficulty levels but requires complex inputs. In contrast, Twee offers an intuitive interface for generating diverse, pedagogically valid tasks, supported by corpus-based authenticity and cognitive validity. The study concludes by stressing the need for further empirical research and encouraging teachers to engage with this emerging technology.

키워드