Africa-wide researchers have made what they explain to be the biggest dataset of African languages for the growth of artificial intelligence known to the public. One of the most significant contributions of the African Next Voices project is the complete restructuring of datasets for 18 African languages to enable AI systems to work with them.
This move by the African Next Voices project is a step toward confronting the issue of the current situation where AI systems are heavily influenced by English and other European languages.
This project over two years period has brought together the expertise of the dentists and computer scientists from Kenya, Nigeria and South Africa. They were recording the sounds for the scenes that portrayed the day-to-day activities of mankind in the fields of agriculture, health, and education.
The team, through their work, went deep into rural lives and recorded the farm scenes, the clinic, and school scenes in Kikuyu and Dholuo for instance in Kenya. They also recorded community activities happening in Hausa and Yoruba in Nigeria and is Zulu and Tshivenda in South Africa.
According to Prof. Vukosi Marivate, who is the head of the South Africa group of researchers at the University of Pretoria, the likes of ChatGPT are examples of present-day AI tools, which are mostly trained on the English language; and if not then on some other languages that have a vast repository of online texts. Many African languages that are primarily oral have quite a few datasets in writing.
The Gates Foundation committed $2.2 million to this cause. This can be used as an open-access data source for developers who create tools that can translate, transcribe, and respond in African languages.
The computational linguist Lilian Wanzare, who was on the Kenyan team, said that they were able to include voices from different geographical locations, and that they were of different ages, and came from different walks of life. Hence they ended up with datasets that were very inclusive. The project covers the languages spoken by millions of people but is just a tiny representation of the 2,000 or so languages estimated to exist on the continent.
