Overview of GRACE at IberLEF 2026: Granular Recognition of Argumentative Clinical Evidence

Resumen

This paper presents the GRACE shared task at IberLEF 2026, the first shared task on clinical Argument Mining in Spanish. GRACE comprises two tracks: Track~1 focuses on argumentative component and relation identification in Randomized Controlled Trial (RCT) abstracts, while Track~2 addresses hierarchical argument mining in clinical cases from the Spanish MIR examination. A total of 16 teams registered, of which six submitted official runs and system description papers. Participants explored diverse approaches, including encoder-based Transformer models, large language models (LLMs), ensemble methods, and zero-shot prompting. The results demonstrate the feasibility of the proposed tasks while highlighting the challenges posed by clinical argument mining.