2026
Authors
Vicent Ahuir Esteve , Marín Muñoz, Sergio , María José Castro Bleda
Publication
Iberian Languages Evaluation Forum
Abstract
The detection of misogyny in song lyrics introduces unique linguistic and cultural challenges, including implicit bias, genre-specific tropes, and contextual ambiguity. In this working note, we report our participation in the MiSonGyny shared task across all three official tracks. We describe our chunkingbased preprocessing pipeline, domain-adapted model selection, and task-specific training strategies. Our system achieves an overall ranking of 9th-10th place, with Task 1, Task 2, and Task 3 Macro-F1 scores of 0.8334, 0.6562, and 0.7724, respectively. We discuss key technical decisions, failure modes, and cross-task generalization patterns, and outline directions for future work in culturally grounded bias detection.
Loading...