ETEL: A Lightweight System for Readability Assessment in Basque

Resumen

Automatic readability assessment is challenging for low-resource languages such as Basque. This paper introduces ETEL, a modular and efficient system for analyzing and classifying Basque texts according to CEFR levels. The system includes a web portal, API, and user management for collaborative corpus creation. ETEL's core uses six computationally inexpensive and interpretable linguistic features. We evaluated these on the 708-text HABE corpus, using statistical analysis to validate their discriminative power and machine learning experiments to benchmark performance. A Random Forest model using ETEL's features achieves 0.83 accuracy in a 3-level (Basic/Intermediate/Advanced) classification and a competitive 0.93 accuracy in a binary (Basic/Advanced) task. This work provides preliminary empirical evidence supporting the effectiveness of ETEL as a lightweight tool for Basque readability assessment, offering a valuable resource for educators and a methodological blueprint for other low-resource languages. The system was developed by two nodes of the CLARIAH-ES infrastructure.