Language learners make the fastest progress when reading texts that match their proficiency level. But most real-world texts are too hard—and manually adapting them is time-consuming. So the big question is: Can AI automatically simplify texts to a specific CEFR level without losing meaning? We explored exactly this in the TSAR 2025 Shared Task, where systems had to rewrite advanced English texts (B2+) to easier levels like A2 or B1. Our team submitted two different approaches: EZ-SCALAR and SAGA.
EZ-SCALAR works like an expert panel of AI models. Two large language models (GPT-5 and Claude) each produce their own simplification, critique each other, refine their versions, and then a final “judge” model picks the best result. An extended version, EZ-SCALAR Lex, adds something extra: a vocabulary check using EFLLex, a CEFR-mapped word list that flags words that are too advanced. This gives the models explicit guidance instead of relying purely on intuition. SAGA, in contrast, uses an iterative proposer–reviewer loop: one model rewrites the text, a classifier checks whether the level is correct, and the proposer keeps revising until the reviewer is satisfied.
So what worked best?
EZ-SCALAR Lex came out on top.
Across the official test metrics, it more reliably hit the target CEFR level while still keeping the meaning of the original text. The vocabulary guidance turned out to be key—external linguistic knowledge gives AI systems clearer boundaries and leads to more controlled simplifications. SAGA, while good at preserving surface similarity, was more conservative and often failed to simplify enough.
Overall, our systems finished in the middle of the leaderboard—but the experiments highlight a bigger takeaway: Combining AI models with structured linguistic resources is more effective than relying on AI alone.
As automatic simplification becomes more important for education and accessibility, hybrid systems like these offer a promising way to create better, level-appropriate texts for learners while keeping meaning intact.
Paper: Alfter, D., & Gombert, S. (2025). GRIPF at TSAR 2025 Shared Task Towards controlled CEFR level simplification with the help of inter-model interactions. In Proceedings of the Fourth Workshop on Text Simplification, Accessibility and Readability (TSAR 2025) (pp. 137-148).