moses-smt/mosesdecoder

Health v2 · maintenance only

Adoption not part of the score

1624 stars · 772 forks observed · 2026-08-28

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

Moses is a statistical machine translation (SMT) system that trains translation models from parallel corpora and decodes with efficient phrase-based and syntax-based search. It includes training pipelines, tuning tools, a fast decoder (plus Moses2), and integration with language models like KenLM.

Use cases

• train a machine translation model from a parallel corpus
• translate text between language pairs with a statistical MT decoder
• build phrase-based or syntax-based translation systems
• tune translation model weights with MERT/MIRA
• run machine translation as a web service
• experiment with factored and sparse feature translation models

When to choose

• you need a classic statistical MT system with full training and decoding pipelines
• you have a parallel corpus and want to train translation models for any language pair
• you need an efficient, customizable decoder with advanced features like factored models and sparse features
• you want to run MT as a server or integrate it into a larger pipeline

When to avoid

• you want modern neural machine translation quality - use NMT frameworks instead
• you need a lightweight library to embed in another application
• you are starting a new MT project, since SMT is largely superseded by neural approaches

Facets

application · maturity maintenance

machine-learning nlp cli http-server machine-learning artificial-intelligence windows cpp cli statistical-machine-translation smt decoder phrase-based-translation language-models kenlm natural-language-processing linux macos

Member repositories

For agents

markdown · JSON · MCP: product_card(name="moses-smt/mosesdecoder")