Overview of NivELE at IberLEF 2026: Automatic leveling of texts written by students of Spanish as a foreign language

Resumen

The automatic classification of language proficiency levels is a crucial diagnostic phase in organizing courses of Spanish as a Foreign Language. In this paper, we present the results of the IberLEF 2026 shared task, NivELE: Automatic leveling of ELE students texts, which focused on predicting levels of the Common European Framework of Reference from written productions of ELE learners. The shared task, organized on Kaggle, attracted 15 registered teams, of which 9 actively participated and 5 submitted their working notes describing their systems. The results demonstrate that fine-tuning pre-trained encoder models in Spanish represents the most robust approach for this task, while also highlighting the challenge of mitigating thematic over-fitting.