Logo Lanfrica

YumiMilling/Liseli

Domain:

natural language processing

Record type:

dataset
Creator:
Yum
Host:
Zambia's Open Language Project — crowdsourcing AI training data for Bemba, Nyanja, Tonga, Lozi, Kaonde, Lunda, and Luvale # Liseli **Zambia's Open Language Project** **Zambia's Open Language Project** **Liseli** ("light" in Lozi) is Zambia's open language project. We're building the dataset that will teach technology to understand Zambian languages — **Bemba, Nyanja, Tonga, Lozi, Kaonde, Lunda, Luvale**, and more to come. If you speak a Zambian language, you can contribute. Translate a word, validate a phrase — every contribution counts. --- ## Why This Exists AI speaks English. It doesn't speak Bemba, Nyanja, or Tonga. The 20 million Zambians who use these languages daily are locked out of AI tools that could transform health, agriculture, education, commerce, and government in Zambia. No tech company will fix this. The market is too small, the languages too niche, and the linguistic expertise too local. **This is infrastructure that Zambians have to build ourselves.** Languages that don't exist digitally lose status. If AI speaks English but not Nyanja, it reinforces that Nyanja is "less modern." Putting Zambian languages in AI is an act of preservation — making them languages of the future, not relics of the past. Liseli crowdsources translations, validates them through community consensus, and produces open datasets that belong to Zambia. The data is CC BY-SA — free, open, owned by the community that built it. ## The Data | Language | Parallel Pairs | Corpus | Words | |----------|---------------|--------|-------| | Nyanja (Chinyanja) | 57,733 | 79,096 | ~1.7M | | Bemba (Icibemba) | 31,785 | 40,104 | ~755K | | Tonga (Chitonga) | 31,751 | 39,897 | ~746K | | Lozi (Silozi) | 31,803 | 43,931 | ~805K | | Luvale | 31,777 | 39,535 | ~830K | | Lunda (Chilunda) | 31,804 | 37,979 | ~691K | | Kaonde (Kikaonde) | 8,888 | 18,228 | ~347K | **Sources:** Storybooks Zambia (CC-BY), Bible translations, MoE teaching modules, USAID Let's Read, ZambeziVoice, FLORES-200, and community contributions. ## How It Works 1. **Translate** — See an English word or sentence. Translate it into your …