Overview of PROFE at IberLEF 2026: Language Proficiency Evaluation

Resumen

The PROFE task evaluates automatic Spanish reading comprehension systems using exercises drawn from real Instituto Cervantes exams, preserving the evaluation conditions of human students and without providing task-specific training data. In this paper, we present the second edition of the task, held within the framework of IberLEF 2026. The main novelty of this edition is the inclusion of exercises with images, which makes it possible to assess the extent to which systems can combine textual and visual information while also deciding when visual cues are relevant or merely decorative. Nine teams registered for the task, and three submitted official runs. The results show strong performance on multiple-choice exercises, notable progress on matching exercises, and greater difficulty in fill-in-the-gap exercises, as well as substantial differences between exercises with and without images depending on the approach used.