About Vocorex

Nothing becomes part of the verified record because one person said so. Every contribution goes through the same process:

1. Independent review. At least two native speakers look at a contribution separately. They cannot see each other’s verdict beforehand.
2. No self-approval. A contributor can never review their own work. This is enforced by the database itself, not only by the application, so it cannot be bypassed by a bug in the code.
3. Disagreement escalates. If reviewers disagree, the item does not quietly go one way or the other. It goes to a senior linguist who resolves it and records why.
4. Corrections, not just rejections. A reviewer can ask for a specific change rather than rejecting outright, and the contributor can revise and resubmit.

Automatically generated content is never treated as verified. If a machine translation model produces something, it is labelled as a draft and must be reviewed by a person before it counts.

Contributors keep ownership of what they contribute. What they grant is permission for specific uses, chosen individually rather than bundled into one accept-everything button:

• Storing written contributions so reviewers can check them.
• Storing voice recordings.
• Including contributions in language datasets.
• Using contributions to train speech and translation models.
• Publishing contributions in an openly downloadable dataset.
• Commercial use of the resulting datasets or models.
• Building a synthetic voice from recordings.
• Showing the contributor’s name publicly.

Only the first is required. Everything else can be declined while still contributing fully. Synthetic voice is off unless a contributor deliberately turns it on, and cannot be enabled on their behalf.

Consent can be withdrawn at any time. When it is, the contribution stops being used for that purpose immediately and is excluded from every dataset built afterwards.

There is one thing we cannot undo, and we say so plainly rather than burying it: if a contribution has already been published in an open dataset, copies other people downloaded cannot be recalled. That is why public release is a separate, optional choice.

It is not a finished translation service. The verified record is small, and it will stay small until enough people contribute to it. Where the platform does not know something, it says so.

Speech recognition and spoken output for Zomi are experimental. There is not yet enough recorded Zomi speech to train a model that could be relied on, and building that corpus is precisely what the recording tools are for. Until then, anything those features produce is labelled as unverified.