Automatic Transformation of Textual Content Using LLM-Technology: a Feasibility Study

This study investigates the feasibility of using Large Language Model (LLM)-technology to classify and transform textual (educational) content according to CEFR proficiency levels in support of digital accessibility in higher education. We developed a browser-based Proof-of-Concept that performs on-the-fly CEFR classification and transform text to target levels (A1–C2) in Dutch and English. Using a corpus of 120 texts evenly distributed across CEFR levels, we evaluated classification accuracy and transformation effectiveness under multiple temperature settings. Results show modest zero-shot classification performance and systematic mid-level bias, particularly for Dutch. Transformation outcomes were stronger in English, especially for B1–B2 targets, but weak for extreme levels and prone to instability. The findings suggest that while LLMs show promise for automated readability adaptation, reliable deployment requires task-specific tuning, multilingual robustness testing, and human-centered evaluation to ensure meaningful accessibility gains.