About
Why this exists
Kenyan languages are documented in dictionaries, academic work and community projects going back decades. Very little of it sits in one place, in a format software can read, under a licence that lets anyone build on it. That is the gap this is trying to close.
I started this because too many Kenyan languages are digitally invisible at exactly the moment language technology is moving fastest. If a language is missing from dictionaries, phrase banks and machine-readable datasets today, it does not simply stay where it is. It falls further behind, and the people who speak it lose access to tools everyone else gets by default.
My role here is not only technical. I am deciding the data model, the moderation rules, what counts as verified and what does not, and how the translation layer should work as it grows past single words. Those choices matter more than the code.
I also build Usalama Voice, an anti-GBV platform for emergency response and survivor support in Kenya. The two are connected. In a crisis, a person needs to understand instructions and describe what happened in the language they trust most, and that only works if the language exists in the system.
If our languages are absent from the systems shaping the future, our communities are absent from that future too
Where it stands
- 4,712
- Words a person has checked
- 36
- Languages
- 0
- Recordings
- 1,557
- Entries held back for review
The corpus was seeded from published sources and is now being checked by speakers. 1,199 entries are held back until someone writes a real definition, and 358 until a speaker confirms the spelling, because an old import damaged it. Held-back entries are excluded from every public count. We would rather publish a smaller number that is true.
Audio collection is just beginning. It is the priority, because recordings are the one part that cannot be gathered later from speakers who are no longer here. How we decide what counts.
What it is for
A dictionary that works on a phone
Search any Kenyan language, in English, Kiswahili, or the language itself, and hear how a word is actually said.
Translation between Kenyan languages
Not only to and from English. Two languages that record a word for the same meaning can reach each other directly.
Training data anyone can use
Licensed CC BY 4.0 so a keyboard maker, a researcher or a speech company can build on it, provided they credit the people it came from.
A record that outlasts its speakers
Some of these languages have few fluent speakers left. Written entries with recorded audio are what survive them.
Who should be doing this with me
It started as a community project and it still is, but the more I look at what it has to become, the clearer it is that community contribution alone will not get there. Reviewing rights are granted per language, so a speaker of one language never has authority over another.
First-language speakers
The final authority on their language. No degree required, and none implied.
Linguists and students
Structure, orthography, and the judgement calls that need training.
Institutions
Universities, archives and language bodies with material already collected.
Engineers
The speech layer, the export formats, and everything after the dictionary.
If you speak one of these languages, you already have what this needs
A word takes about a minute. A recording takes ten seconds. If you work in language, research or education and want to do something larger, get in touch.