Overview of MSLG-SPA at IberLEF 2026: Bidirectional Translation between Mexican Sign Language Glosses and Spanish

Resumen

This paper presents an overview of the MSLG-SPA shared task at IberLEF 2026, the first shared evaluation campaign devoted to bidirectional translation between Mexican Sign Language glosses (MSLG) and Spanish. The task comprised two subtasks: Gloss-to-Spanish and Spanish-to-Gloss translation. Twenty teams submitted a total of 100 system runs, exploring a wide range of approaches including multilingual encoder-decoder models, instruction-tuned large language models, parameter-efficient fine-tuning, retrieval-augmented generation, and hybrid rule-based systems. Systems were evaluated using BLEU, METEOR, chrF, and COMET, together with a standardized multi-metric ranking strategy. Beyond presenting the official evaluation results, this overview analyzes the methodological trends observed across the participating systems and establishes one of the first public benchmarks for Mexican Sign Language translation, providing resources to support future research on low-resource sign language processing and accessibility technologies.