DEPART: DEcomposing PARiTy across Multilingual LLMs
Published in EMNLP 2026 (Findings), 2026
A framework for decomposing the sources of cross-lingual performance gaps in multilingual LLMs. Paper: arXiv:2605.28163.
Manan Uppadhyay
Research Fellow, Microsoft Research India. Multilingual NLP, interpretability, safety.
- Bangalore, India
- Microsoft Research India
- Google Scholar
- GitHub
- X (formerly Twitter)