AI-based Graded Texts for Language Learning

Overview

Reading graded texts — passages written or adapted to match a learner's current proficiency — is one of the most effective methods for acquiring vocabulary and grammar in context. This project uses language models to generate, adapt, and automatically grade reading passages according to established proficiency frameworks (e.g., CEFR A1–C2), removing the manual bottleneck of content creation for educators.

How It Works

1. 1A topic and target proficiency level are selected by the user or educator.
2. 2An LLM generates a reading passage constrained to vocabulary and syntactic complexity appropriate for that level.
3. 3A grading model scores the generated text against readability and proficiency metrics.
4. 4If the score falls outside the target band, the passage is revised automatically.
5. 5The final text is presented with comprehension questions and vocabulary highlights.

Grading Approach

Placeholder — describe the readability metrics used (e.g., Flesch-Kincaid, CEFR word lists, syntactic complexity features) and whether the grader is a fine-tuned classifier, a prompted LLM, or a rule-based scoring pipeline.

Results & Evaluation

Placeholder for evaluation results — inter-rater agreement between the automatic grader and human annotators, user study outcomes, or accuracy on a labelled proficiency dataset. Add charts or example outputs here.