752
Tests
0
Dependencies
5
Script converters
47
Language registers
Mix methods per language pair
Each source→target pair gets its own translation method, model, and quality configuration. Use what works for each language — not a one-size-fits-all.
Learn about methods →{
"version": 3,
"pairs": {
"en:fr": {
"method": "google-translate"
},
"en:ja": {
"method": "llm",
"model": "google/gemini-2.5-pro"
},
"en:crk": {
"methodPlugin": "crk-coached-v1"
}
}
}
Built For
SaaS Internationalization
Translate your Next.js, Hugo, or React app to 30+ languages with per-pair quality control.
Translate 30 Languages →Build a Custom Pipeline
Chain LLMs with FST validators, dictionaries, and post-processors. Package it as a plugin.
FST Pipeline Cookbook →Language Preservation
Coached LLM translation for languages with no API coverage — Indigenous, endangered, constructed.
Low-Resource Guide →THE ARENA
Think you have the best method
for translating English → Plains Cree?
Prove it.
7,000+ languages. ~130 have machine translation. The rest are an unsolved problem — and an open invitation. rosetta's evaluation harness benchmarks any method with fingerprinted, reproducible scoring. The leaderboard tracks every submission.
Plug and Test
Run your method against standardized benchmarks. chrF++, exact match, FST acceptance — all computed by the same harness.
Eval Harness →Claim Your Score
Every submission is fingerprinted to a Git commit and scored against the same dataset. Open a PR to submit.
Leaderboard →Respect the Data
Indigenous languages belong to their communities. rosetta supports OCAP, CARE, and Māori Data Sovereignty principles.
Data Sovereignty →This is an unsolved problem that everyone in the world can contribute to.
Build a method. Score it. Give it back.