Scaling Neural Machine Translation to 200 Languages

Introduction

1. What problem does No Language Left Behind (NLLB-200) address?
   Neural MT quality depends on large parallel corpora, which exist for few of the world’s 7,000+ languages, so progress has mostly benefited high-resource languages and widened digital inequities. NLLB-200 is a single model translating among 200 languages, three times as many low- as high-resource.
   • Team, 2024, Scaling neural machine translation to 200 languages (Nature, vol. 630), Introduction, p. 841
2. How does NLLB define a high-resource language?
   One with at least 1 million sentences of aligned text (bitext) with another language; a very low-resource language has fewer than 100,000 samples across all its pairings.
   • Team, 2024, Scaling neural machine translation to 200 languages (Nature, vol. 630), Introduction, p. 841
3. What overall improvement did NLLB-200 achieve?
   An average 44% improvement in translation quality (BLEU-based) over the previous state of the art, while doubling the number of supported languages, showing that wide coverage and quality aren’t mutually exclusive.
   • Team, 2024, Scaling neural machine translation to 200 languages (Nature, vol. 630), Introduction, p. 841
4. What three evaluation tools did the NLLB project create?
   FLORES-200, a benchmark covering about 40,000 translation directions; XSTS, a human evaluation protocol focused on meaning that is consistent across languages; and ETOX, a detector of toxicity added by translation, covering every language.
   • Team, 2024, Scaling neural machine translation to 200 languages (Nature, vol. 630), Introduction, p. 842