Meta's Omnilingual MT system enables machine translation for 1,600 languages using new datasets, LLaMA3-based models, and evaluation tools; outperforms 70B LLM baseline with smaller 1B-8B models.

Omnilingual MT: Machine Translation for 1,600 Languages | Research

RESEARCHNLPMarch 17, 2026AbstractAdvances made through No Language Left Behind (NLLB) have demonstrated that high-quality machine translation (MT) scale to 200 languages. Later Large Language Models (LLMs) have been adopted for MT, increasing in quality but not necessarily extending language coverage. Current systems remain constrained by limited coverage and a persistent generation bottleneck: while crosslingual transfer enables models to somehow understand many undersupported languages, they o...