From Gutenberg to the classroom: large-scale generation and validation of vocabulary-controlled EFL reading materials
Yujong Park, Ali Derakhshan
Abstract
Developing level-appropriate, story-based reading materials for EFL contexts is a labor-intensive process that often exceeds the time constraints of classroom teachers. This study provides a preliminary validation of an automated pipeline designed to bridge the materials gap by transforming public-domain children’s literature into curriculum-aligned instructional resources. Using GPT-3.5-turbo, we processed 1,574 source texts from Project Gutenberg, systematically simplifying them to match the Korean Ministry of Education’s 3,000-word vocabulary list. The resulting corpus contains 116,809 passages, each featuring six machine-generated comprehension questions and thematic classification via Latent Dirichlet Allocation (LDA). To test the pedagogical viability of the materials, a sample of 23 Korean university students participated in our study. Participants achieved an 85.7% accuracy rate on comprehension tasks, reporting higher engagement with the narrative-driven content than with traditional textbook excerpts. However, qualitative analysis revealed that while the AI achieved high vocabulary precision, human oversight remains necessary to correct occasional disruptions in narrative coherence and cultural nuance. The results suggest that while AI can reduce material development time, teachers should exercise high-level oversight to ensure sociocultural and narrative integrity. By releasing the complete pipeline and dataset as open-source resources, this research suggests one model for teacher-AI collaboration in materials development.