Meta AI NLLB-200 Multilingual Translation – AI & NMT RESEARCH

Meta AI’s NLLB-200 (‘No Language Left Behind’): Open-Sourcing a 200-Language Neural MT Model

In July 2022, Meta AI unveiled NLLB-200 (‘No Language Left Behind’), a single open-source 54.5-billion-parameter Mixture-of-Experts model capable of translating directly across 200 languages—including 55 African languages—backed by the FLORES-200 evaluation benchmark.

Advancing Low-Resource Machine Translation by 44% on FLORES-200

In July 2022, Meta AI published its landmark research project ‘No Language Left Behind’ (NLLB-200), releasing both a 54.5-billion-parameter Sparsely Gated Mixture-of-Experts (MoE) model and smaller distilled variants (600M, 1.3B, and 3.3B parameters) under an open license. Capable of translating across 200 languages (40,000 translation directions), NLLB-200 improved translation quality by an average of 44% in BLEU/spBLEU scores over previous state-of-the-art systems across the newly constructed FLORES-200 human-translated benchmark.

Meta AI NLLB-200 Multilingual Translation – AI & NMT RESEARCH
AI & NMT RESEARCH — Professional Localization & Regulatory Quality Assurance (July 11, 2022)

Evaluating 40,000 Language Directions with Human-Curated Benchmarks

To ensure reliable evaluation across low-resource languages—including 55 African languages such as Luganda, Igbo, Wolof, and Zulu—Meta collaborated with professional translators and native linguistic reviewers to create FLORES-200, a parallel evaluation dataset where every sentence is human-translated into all 200 languages. Additionally, NLLB-200 introduced toxicity filtering lists across 200 languages to prevent machine translation hallucinations from introducing offensive terms.

Executive Takeaways for Global Enterprises

  • Key Insight: Meta AI’s NLLB-200 supports direct translation across 200 languages, achieving a 44% average quality improvement on low-resource pairs.
  • Key Insight: The open-source FLORES-200 dataset provides professionally translated human evaluation benchmarks across all 40,000 language directions.
  • Key Insight: Distilled NLLB models (1.3B and 3.3B) enable researchers and LSPs to fine-tune domain-specific engines for underserved languages.

Hai bisogno di traduzioni multilingue certificate, asseverazioni legali o localizzazione conforme agli standard ISO per i tuoi progetti internazionali? Richiedi un preventivo gratuito personalizzato ad Alphatrad e ricevi una risposta dettagliata entro 24 ore.


Editorial & Copyright Attribution Note: This industry analysis has been independently synthesized and adapted by Alphatrad’s editorial team based on verified regulatory documentation and industry research originally published by Meta AI Research. For the original publication and primary documentation, please consult:
Source: https://ai.meta.com/blog/nllb-200-high-quality-machine-translation/

0 risposte

Lascia una risposta

Vuoi partecipare alla discussione?
Apporta il tuo contributo!

Lascia un Commento

L'indirizzo email non verrà pubblicato. I campi obbligatori sono contrassegnati *

15 − quattordici =