Automated Item Generation for Reading Assessments

Le résumé fourni par la source

This study explores the application of automated item generation (AIG) using large language models (LLM) within the PIRLS (Progress in International Reading Literacy Study) reading assessment framework. By operationalizing PIRLS item-writing guidelines via GPT-based prompting strategies, the study demonstrates a scalable approach to developing reading-comprehension items. The item generation process involved two main steps: first, generating passage-specific items using structured prompts; second, filtering items using similarity metrics. After this automated filtering, experts reviewed the remaining items to ensure content quality and alignment with PIRLS standards and curated the set. As a final step, the items were evaluated psychometrically against human-authored items. The results suggest that LLM-based AIG can be a valuable tool to support item development in large-scale international reading assessments.

Ce résumé expose les affirmations des auteurs. BNTIC ne l’interprète pas comme une validation indépendante des résultats.