What we believe
Community-contributed corpora carry do-not-train flags and the licenses their owners choose. We follow OCAP® (Ownership, Control, Access, Possession) and CARE principles — they're design constraints here, not a press release.
If a translation method only works because speakers shared their grammar, their dictionaries, or their time, the community has a standing claim on it. The ownership-transfer documentation spells out how that is meant to work.
Each language card records where every fact came from — and that means every fact can be challenged and corrected by the people who actually speak the language. You are the authority on your language; we are not.
How to start
- 1Look up your language's card
See what the public record says about your language — names, vitality, scripts, resources — and what it gets wrong. Find your language →
- 2Tell us what's wrong (or missing)
Corrections from speakers outrank any database we ingest. Open an issue on GitHub — no technical background needed, plain words are perfect.
- 3Shape what gets built
Benchmarks for your language should be reviewed by people who speak it. If you want your community involved in that review — on your terms — start with the community documentation below and reach out.
Contests and community programs are in development. We won't announce them here until they're real.