Why we exist
Logha Labs is making AI work in under-resourced languages.
Models learn from the internet, and most of the world’s languages barely appear on it. We build native data and expert evaluations for languages with a limited online presence, so AI works as well for the people who speak them as it does in English.
loghaلغة
Arabic, noun. Language.
The same word, in some of the languages that have the least written about them online. Swahili and Somali borrowed theirs from Arabic.
- lughaSwahili
- ቋንቋAmharic
- èdèYoruba
- harsheHausa
- ভাষাBengali
- மொழிTamil
- भाषाHindi
- زبانPersian
- ژبهPashto
- ភាសាKhmer
- luqadSomali
- asụsụIgbo
- ulimiZulu
- làkkWolof
- simiQuechua
- wikaTagalog
The gap
English gets the best AI. Everyone else gets the rest.
Models learn mostly from the web. Languages with a small online footprint get weaker models, and their dialects and specialist fields get weaker still. The data and the experts aren’t there yet.
Our mission
Make under-resourced languages as powerful as English for LLMs.
لكلّ لغةٍ مكانٌ في المكتبة
- NowArabicModern Standard and the major dialects.
- NextMore under-resourced languagesWidely spoken, rarely written online.
In the bindery · under construction
Benchmarks & the state of Arabic data
How today’s models really perform in Arabic, by dialect and domain.
For experts
Lend your expertise to the collection.
Flexible work and strong pay for tasks within your domain. Help advance AI models beyond English, and earn flexibly and remotely.
Apply as an expertBuilding for Arabic speakers? Let’s talk data.
Tell us the language, dialect and domain. We’ll come back with a sample and a plan.
Talk to our team