facebookresearch/UnsupervisedMT

Health v2 · maintenance only

Flags: no_releases archived no_license

Adoption not part of the score

1499 stars · 261 forks observed · 2026-08-28

What it is AI-extracted, prompt v1, taxonomy v1, 2026-08-30, confidence not recorded

The original implementation of phrase-based (PBSMT) and neural (NMT) unsupervised machine translation models from the EMNLP 2018 paper by Facebook AI Research. It supports seq2seq, biLSTM+attention, and Transformer architectures with denoising auto-encoder and back-translation training.

Use cases

• translate text between languages without parallel training data
• train an unsupervised neural machine translation model
• run unsupervised phrase-based SMT with Moses
• reproduce the EMNLP 2018 unsupervised MT paper results
• train a Transformer translation model with back-translation
• generate cross-lingual embeddings for translation

When to choose

• you need to reproduce the original unsupervised MT paper's PBSMT or NMT results
• you want a research reference implementation of denoising auto-encoder and back-translation training
• you need unsupervised phrase-table generation with Moses

When to avoid

• you want state-of-the-art unsupervised NMT - use facebookresearch/XLM instead as the authors recommend
• you need a production translation system or maintained library
• you need recent PyTorch compatibility - the code was tested on PyTorch 0.5
• you want a simple pretrained translation model without training infrastructure

Facets

library · maturity maintenance

machine-learning nlp deep-learning machine-learning deep-learning python cli machine-translation unsupervised-learning nmt pbsmt pytorch research-code facebook-research natural-language-processing linux

Member repositories

For agents

markdown · JSON · MCP: product_card(name="facebookresearch/UnsupervisedMT")