AI-driven Vulavula to champion African languages

Simnikiwe Mzekandaba
By Simnikiwe Mzekandaba, IT in government editor
Johannesburg, 17 Mar 2023
Lelapa AI CEO and co-founder Pelonomi Moiloa. (Photograph by Strike A Pose Studios)
Lelapa AI CEO and co-founder Pelonomi Moiloa. (Photograph by Strike A Pose Studios)

Newly-opened artificial intelligence (AI) research and product lab Lelapa AI will go to market with its first natural language model solution, Vulavula, within the next two months.

So says co-founder and CEO Pelonomi Moiloa, explaining the start-up’s plans to build AI-based customer-facing products and services that solve uniquely-African problems.

This comes as the world is witnessing accelerated interest in natural language processing tools driven by AI, with tech giants Google, Microsoft and Meta adding AI into their product offerings and services.

Moiloa tells ITWeb that because of its “very strong base and experience” in the language space, Lelapa AI started with a language product as its first offering.

Vulavula is powered by language model technology, which has received lots of attention since the release of ChatGPT, she explains. “Vulavula is a language technology-as-a-service, focusing on under-represented languages (specifically South African languages), through text and voice technologies.

“Despite 90% of the internet being in English, only one in 10 South Africans speak English at home. Languages represent culture and people, and it’s imperative that people from across the world are supported by technology.”

Moiloa reveals Vulavula has several primary features or capabilities. “The first one is called named entity recognition. This means the machine is able to recognise pronouns – names, cities, locations, numbers, etc. For example, when you type your name into Microsoft Word and it puts a red squiggly line underneath. This kind of technology is able to identify African names as people, or African places as places.

“The next feature is speech to text or text to speech. This is useful for things like subtitling for news articles or TV programmes. There is also a feature for machine translation, which is the most difficult task for a machine to do because it has to understand the language quite well.”

African AI for all

According to Moiloa, Lelapa AI was founded out of the need to address how AI can be used for solutions and applications from an African lens.

The founding members consist of Moiloa, Jade Abbott, Vukosi Marivate, Benjamin Rosman, Pravesh Ranchod and George Konidaris, who have backgrounds in academics, research, data science and engineering.

Moiloa explains that in 2018, co-founder Abbott realised there was no research that existed around machine understanding of local languages on the continent. Abbott wanted to learn isiZulu and could not find a technical tool to help her do that.

“She set out to develop this grassroots movement, because she knew she couldn’t do it alone, to develop the research field which understands African languages from a contemporary machine methodology.

“This research institution contributes to understanding how to better adapt technologies that exist to our local context and local languages, so that machines are able to better understand us and facilitate us communicating in our languages through certain tools.”

She highlights that the team hopes to make a change in the country, using AI to improve sectors such as agriculture, education, healthcare and energy.

“Africa has many unique issues that could be addressed by AI solutions, and these issues need to be addressed by people who understand the nuances of the context, rather than by Western counterparts. When other parts of the world aim to solve our problems, they often fail because of this lack of context − and quite often cause harm.

“Right now, there is a wave of AI talent on the African continent − many of these are PhD graduates who left the continent to study abroad. Unfortunately, despite the many problems they could be solving, that talent has very few places to go, and so often choose to remain abroad or end up working somewhere that can’t take advantage of their skillset. Lelapa AI aims to be the home of that talent.”

The co-founder indicates the AI lab’s products are focused on the business-to-business market, as these companies have to communicate with their customers on a regular basis. This is mainly communication through text or voice.

“For example, in a text-based application like any of the banking apps that communicate through a chat application – for payments, ordering airtime or prepaid electricity via text.

“From an audio perspective, these are businesses that have to deal with call centres, or any type of service industry where you’re calling in because you’re trying to get something done and you have to speak to a person to sort through an issue.”

From a commercial perspective, she notes Vulavula is available predominantly in isiZulu for now. “However, we’ve already done a lot of the ground research in order to support the next languages.

“By adding isiZulu, Xhosa, Sesotho, Setswana and Afrikaans, we get up to 80% of the population and create access for people to be spoken to in their own languages and speak in their own languages.”

Moiloa concludes that the lab would like to expand its product offerings into other fields, but right now, the focus is language and that product is Vulavula.