For Data and Community Partners
Language communities, institutions and reviewers who help build African-language data on fair terms.
Language Communities
A group that joins together can hold a shared agreement and share in the community fund. Think of a village association, a church choir or a university department.
No language is given a special place. Ideas to record are designed to fill gaps: under-represented languages, dialects, topics and speakers.
The Community Fund
A share of licence income is planned for a community fund. It would pay for language-community projects, governed under the CARE principles.
The size of each share is to be decided. No benefit-sharing agreement exists yet. Who governs the library and the fund, a trust, a cooperative or a foundation, is also to be decided.
Benefit-sharing and governance are tracked as sovereignty indicators.
Institutions
Universities, research groups and data holders can help in three ways.
Every source gets a licence record. No text enters training without one.
Reviewers
Native speakers review recordings and transcripts. In the plan, reviewers are paid per item and drawn from members with long good records.
The tokenizer has no native-speaker reviewer yet for any of its 12 Tier 1 languages. Reviewers for all 12 are the target.
Open Needs
No open Arabic-script training text exists for Tachelhit or Hassaniya. They are the only two of 26 languages where Nzinga needs more tokens than the best other tokenizer: 2.6% and 5.3% more.
Today no training text comes with community consent or from African partners. The target is a majority.
To offer data or join as a community, contact the team.