The French Development Agency (AFD) organized a conference aimed at exploring the challenges of developing innovative artificial intelligence (AI) models and technology while taking into account the linguistic specificities of African languages. 

This gave the panelists, led by Peter Addo, Head of Emerging Tech Lab at the AFD and people who attended the online event, an opportunity to discuss the current state of AI in Africa bearing in mind the extraordinary linguistic diversity of the continent. Challenges in the domain of data access in African languages as well as ways of further developing AI in an appropriate and well adapted way in the region were also discussed. 

With over 3000 languages spoken in Africa and a significant part of the almost 1.5 billion people living on the continent not speaking either English or French, it is important that AI is developed with such diversity and specificities at its core.

“Chat GPT for example knows the capital of Kenya but when being asked in Kinyarwanda (one of the official languages of Rwanda), it does not understand the question. If we do not put in place the necessary requirements, we risk having a huge gap where people on the continent who do not speak foreign languages like English and French do not get access to AI services. 13% of languages on the continent are endangered and it is still a growing number. As digital transformation takes roots in Africa, they will have to adapt to other languages in order to be able to use those services. AI could play a role in lessening the importance of local languages for youth, for instance, who will be more likely to learn English or French. In 10 or 15 years of digital transformation, it is possible to witness the suppression of several of these local languages, not spoken in online interactions,” explains Audace Niyonkuru, CEO of  Digital Umuganda, an Artificial intelligence company specializing in language technology for African languages with an emphasis on voice based solutions. 

The practice of “Umuganda”, a community-driven practice aimed at promoting social and environmental growth, was the inspiration behind the creation of the digital Umuganda to spark collective action.  The word Umuganda can be translated as ‘coming together in common purpose to achieve an outcome’. In traditional Rwandan culture, members of the community would call upon their family, friends and neighbors to help them complete a difficult task. Now in Rwanda, on the last Saturday of each month, everyone works together to improve their communities as part of Umuganda.

“We told ourselves that if this worked for physical infrastructure then we should try it for digital infrastructure because it is something everyone understands. We also needed diversity from our contributors so we looked for something that united the physical and digital spheres. We then started Digital Umuganda to build an open data set. We held training sessions for contributors to avoid bias as well,” Audace explains. 

“We worked more with communities who understood that the outcome would be preserving the language. The contributors were interested in preserving Kinyarwanda. It was in a framework people understand and are used to, it was self-explanatory to them. It was a voluntary contribution for an infrastructure that would be public afterwards,” he added.

The limitations or possible hurdles on the journey to facilitating AI use by speakers of local African languages were not ignored during this conference as explained by Audace Niyonkuru. 
“AI could be a tool for localisation,” Audace explains, but also a way to facilitate the socio-economic development of the continent by facilitating trade and interactions. 

“Translations between Wolof and Kinyarwanda could help commerce between Senegal and Rwanda for instance, but we also need to understand that some languages are spoken but not written and machine languages are domain-specific. You would have to have a specific data set of terminologies depending on the domain.”

Audace pointed out the lack of data privacy laws in most African countries which further complicates the use and development of “fair” AI: 

“This causes a lot of problems because you have innovators that are being denied data that should be openly available and data by users that are being exploited in a way that can cause harm. We can also create harm for the users who shared the data without consent but people need to understand this consent in the first place. Data governance should be there but the understanding of the data principles in a way that reflects the reality of the Global South, inscribed in the human rights frameworks should also be considered. It is important to embed local values in the creation of the data sets. For this, we need capacity development of the participants involved. Not only access to the machines, but also looking at stopping brain drain from the continent.”