AI AND LOCAL AFRICAN LANGUAGES

Nowadays, language as a social, economic and technological barrier is increasingly discussed. In the African context, there has been a proliferation of initiatives aimed at making people's daily lives easier. In many countries, the limited mastery of official languages is a reality that has motivated researchers (scientists, sociologists, linguists, etc.) to develop tools to bridge the gap in communication and in the use of modern technology (smartphone, computer, website/platform, etc.). Other justifications are also put forward by researchers, in particular the preservation of languages as a cultural identity and a vehicle for social integration (African Language Program, 2021).English-speaking countries seem to be ahead of the game when it comes to using Artificial Intelligence (AI) technology in language practice, where several major technological innovations have seen the light of day. In 2021, South African telecoms company Telkom and start-up Enlabeler launched the Izwe.ai platform, which offers transcription and translation services in several local languages with the ability to interpret South African accents. This AI-based innovation aims to have a significant impact on the education, health and business sectors in order to reduce inequalities. In Rwanda, Software Development Engineer Remy Muhire was part of a team that created a new open-access speech dataset for the Kinyarwanda language, which involves many volunteers registering in the local language: Kinyarwanda. In Nigeria, ARIKPO and DICKSON (2018) have developed an automatic translator from English into the local Nigerian language Efik based on Natural Language Processing (NLP), a branch of AI. In Nigeria, the ARIKPO and DICKSON translator has been preceded by a number of similar tools, albeit based on different methods. Awofolu and Malita (2002) set up an automatic translator from English into Yoruba and vice versa using classic syntactic and semantic analysis algorithms. Folajimi and Omonayin (2012) developed a translation mechanism based on the algorithmic system (Statistical Machine Translation, SMT) that translates from English into Yoruba. Odejobi et al (2006) and Afolabi et al (2013) carried out work on the development of a Yoruba text-to-speech system using the concatenation method. Afolabi et al (2013) state that the interest in applying AI methods to the Yoruba language can be explained by its importance in Nigeria and in several other African countries such as Benin and Togo. The translation technologies developed must be accessible and easy to use. According to the same source, 70% of respondents recognised its ease of use.  

In addition to meeting this challenge of ease of use, recent platforms are attempting to make translation into several languages possible. This has been achieved by the innovative Mashakane platform (Iroro et al., 2020), which offers translations in nearly thirty local African languages. Developed by around twenty researchers (scientists and computer scientists) of African origin, this platform appears to be the most advanced in terms of translation to address development issues in Africa.  

More recently, the search for solutions to take advantage of local languages has been of interest to two students. Dossou (2022) proposed an AI model to help translate Fon into French. His fellow student, Emezue (2022), has been working on translating Igbo into English. Their collaboration resulted in the creation of a translation model called Fon-French Neural Machine Translation (FFR), similar to Google Translate. This innovation makes up for the shortcomings of the current translation engine for local African languages.

Furthermore, the existing literature on the use of AI to support African languages highlights the potential impact on the lives of local populations. In the field of education, UNESCO is promoting AI as an effective means that could be mobilised to meet the challenges and issues facing this sector. The use of local languages in schools is said to contribute to better performance (Ouane and Glanz, 2010).

In Agriculture, AI experiments in Kenya provide clear evidence of the benefits of applying AI for African languages (Brandusescu et al., 2017 and Assefa, 2018). Largely illiterate rural populations will be able to make farm management systems more efficient and resilient, have more loans available through better financial inclusion and build adequate and sustainable infrastructure.

In the field of healthcare, innovative solutions exist and should be generalised to facilitate access to healthcare. Brandusescu et al (2017) and Smith and Neupane (2018) highlight the opportunities of AI to fill the gap in qualified healthcare staff. Having practical advice from healthcare experts in local languages via mobile platforms or apps is a significant advance in medicine especially in rural areas where the availability of healthcare services and infrastructure is sorely lacking.

In the above-mentioned areas, the application of AI is proving to be a necessity in order to overcome the obstacles created by the use of official languages. These obstacles can be strictly considered as inequalities, as a sub-population often finds itself excluded. This exclusion often concerns women, who are increasingly taken into account in the outlook. The development of AI technologies incorporates this gender dimension, which could encourage and accelerate women's entrepreneurship and leadership. Over the past five years, the results on promotion in Sub-Saharan Africa have been satisfactory: in 2017, 27% of entrepreneurs were women, the highest ratio of any region in the world (the Mastercard Index of Women Entrepreneurship, MIWE). AI, by reducing language barriers, is a real opportunity to be seized in the business world.

The development of AI tools for African languages faces many constraints. A number of factors need to be addressed, the most frequently mentioned being the need to improve the legal and regulatory framework and the lack of initiatives to secure funding commensurate with the opportunities that AI can offer.

On a continental scale, Africa has few regional initiatives in terms of AI development and framework policies reflecting the enormous progress to be commonly undertaken (Gwagwa et al., 2020). The flagship initiative is the Convention on Cybersecurity and the Protection of Personal Data (African Union, 2014). At the national level, only 17 of the 55 African Union (AU) member states have adopted comprehensive "data protection and privacy" legislation according to the 2019 Global Information Society Watch African AI Policy Survey. These regulatory frameworks need to be reviewed and updated to take into account the developments observed in the field of AI (Gwagwa et al., 2020).

According to The Web Foundation's 2017 report, the lack of a comprehensive regulatory framework is holding back the development of the sector somewhat. Businesses lack strategic direction and guidance, and the catalytic role of public policy is far from being established. There is a need for coordinated action involving all the players. In addition, the lack of evidence in the field of AI makes evidence based on the use of data difficult if not virtually impossible (Web Foundation, 2017).

Beyond the existence of a still rudimentary legal framework, the low use of local African languages in the global economy dominated by foreign languages (English, French, Chinese, etc.) reduces interest in these languages. The AI giants, motivated by economic profitability, will not invest if real prospects do not follow. This makes it even more difficult to apply AI to local African languages, which then remain dependent on local initiatives set up by the private or state sectors, which are already financially limited by socio-demographic problems (poverty, unemployment, health, etc.). Thus, in Africa, the lack of funding, investment in STEM (science, technology, engineering and mathematics) and the establishment of quality infrastructure are drastically handicapping the possibilities of a strong and inclusive AI industry.