Overview of MentalRiskES at IberLEF 2026: Early Detection of Mental Disorders Risk in Spanish

Resumen

This paper presents the MentalRiskES task at IberLEF 2026, held as part of the 43rd International Conference of the Spanish Society for Natural Language Processing (SEPLN 2026). The goal of this shared task is to advance the early detection of emotional problems in Spanish. We propose two novel tasks: Task 1 focuses on the early detection of symptoms in therapeutic conversations, while Task 2 addresses the selection of the most appropriate therapist response. In addition, participants were asked to report the carbon emissions generated by their systems, emphasizing the importance of sustainable natural language processing practices. In this fourth edition, 30 teams registered, 17 submitted predictions, and 11 submitted system description papers. Most participants experimented with large language model-based approaches, primarily relying on zero-shot prompting strategies with both open-source and commercial models. For Task 1, the best system achieved a combined MAE of 0.879. Task 2 proved considerably more challenging, with the best Macro-F1 of 0.392, only marginally above the random baseline.