Frequently asked

It doesn’t pretend to. An AI can’t truly know a language with fifty speakers, so Lantern never asks it to. The model reasons only over the real phrases it’s given, cites every word to the phrase that attests it, and is forbidden — in code— from ever teaching a word no speaker said. It amplifies the community’s knowledge; it never invents.

It’s a working app, and the honesty is measurable: 0 invented words have ever reached a learner, 48/48 vocabulary items are cited, and 7/7 practice sentences pass attestation. You can recompute the numbers yourself at /api/metrics.

Verified seed corpora, each cited — the Te Aka Māori Dictionary, DAILP and the Cherokee Nation Language Department, Omniglot, and the UNESCO Atlas. New phrases come from contributors and are attributed to them.

Yes — that’s the whole flywheel. The Contribute tab lets anyone add a phrase, and the engine re-derives a richer course in seconds. Every phrase makes the course better. Try it on Māori.

The community does. Endangered-language data carries real ownership and governance concerns, so Lantern treats the community — not the model — as the owner, cites every source, and is building formal data-sovereignty controls so each community governs its own words.

No account needed, and it’s free. You can browse, learn, hear pronunciation, and contribute fully signed-out. Signing in (Google, GitHub, or email) only adds saved spaced-repetition progress and credits your contributions.

In-browser text-to-speech gives a spoken guide using the closest available voice — honest about being a guide, since fluent voices rarely exist for these languages. Speaker-recorded audio from real fluent speakers is rolling out so you can hear the genuine sound.

Eight endangered languages are aboard the Living Ark, with Māori and Cherokee fully learnable today — including Ainu (about ten speakers left), and Manx and Cornish, both declared gone and brought back by their own communities.