Case Study: African Languages and Natural Language Processing (NLP): Challenges and opportunities for civil society

Of the 7,000 languages spoken worldwide, Africa is home to a third of them, making it the most linguistically diverse continent.¹⁵³ Despite this diversity, African languages are under-represented in Natural Language processing (NLP), a subfield of artificial intelligence and computational linguistics concerned with enabling computers to understand, generate, and manipulate human language.¹⁵⁴

NLP techniques are fundamental to the digital tools and platforms we use daily. They are deployed to detect spam, predict what we are typing, moderate content we post online, recognise our speech and convert it into text or to perform an action, for example, with voice assistants. Large Language Models (LLMs) deploy NLP techniques, and they represent a significant advancement in the field. They are trained in large amounts of datasets and perform a wide range of tasks.¹⁵⁵ They power Generative AI models and are a key component of the content moderation operations of social media platforms.

High quality datasets that are free from biases, inaccuracies, and errors are essential in the field of NLP, and large language datasets are necessary for the development of LLMs. These datasets are needed for the training of machine learning models, and when high quality data sets are scarce, there are wide implications for people and their human rights. For instance, their content can be censored because a model misunderstood it to be a violation of platform policies or harmful content. Advocates for African languages, digital rights, and accessibility understand the implications this has in content moderation.

“We have a lot of people on these social media platforms who are communicating in local languages. And so, you find that some of the content that is in local languages cannot be detected by the platforms… Moderation systems will also not address content that is in local languages. Sometimes you find harmful content being disseminated widely or being amplified by algorithms. The lack of investment in training the models in local languages means that users will be affected”, said Victor Kapiyo, senior researcher at CIPESA.

A number of factors are behind the under-representation of African languages in NLP, including a colonial past that marginalised African languages,¹⁵⁶ a painful legacy that post-colonial rulers continued by retaining European languages as official or semi-official languages, and not taking adequate measures to preserve local languages.¹⁵⁷ NLP systems were also not built to meet the needs of many African languages which are structurally different from English, a dominant high-resource language in NLP,¹⁵⁸ or have various substantial dialectal variations. Another challenge is the lack of access to modern computing power and cloud services necessary for training models.¹⁵⁹

All these obstacles exclude African languages from the NLP field. As researchers on the challenges and opportunities of NLP in low-resourced Senegalese languages noted: “Despite their societal importance and widespread use in daily communication and media, these languages remain largely excluded from the digital and scientific landscape of NLP. This gap poses a dual challenge: the risk of technological marginalisation of major linguistic communities, and the missed opportunity to harness NLP for advancing locally grounded research, especially in the social sciences”.¹⁶⁰

The African Union has recognised the language challenge in its Continental African Intelligence Strategy, which includes high level recommendations to promote investment in NLP in Indigenous and local African languages.¹⁶¹ The Strategy’s fifth guiding principle ‘Inclusion and Diversity’ states that “the production, development, use and assessment of Al in Africa will be inclusive, non-discriminatory, leaving no one and no place behind and benefiting everyone, and respectful of the diversity of African people, cultures, languages, gender dimensions and nations”.

To address this gap, civil society and academic initiatives have been launched across the continent. Masakhane, “a grassroots organisation whose mission is to strengthen and spur NLP research in African languages, for Africans, by Africans”¹⁶² has been building open and high-quality data sets that support African language. This includes text and voice datasets in East African languages to support “building natural language processing models that range from spell and grammar checking, sentiment analysis, topic modeling, text summarization, misinformation and fake news detection, machine translation and automatic speech recognition models”.¹⁶³ Another project by the Maskhane community resulted in “the first large, publicly available, high-quality dataset for named entity recognition (NER) in ten African languages”.¹⁶⁴ An important task in NLP, NER is not only essential for automated products and functions like spell-checking and voice assistant but “it also enables identifying African names, places, and organizations for information retrieval”.¹⁶⁵

In another initiative, Code for Africa runs an AI Sandbox with its self-hosted LLMs allowing partners to prototype and experiment with AI models in a secure environment and with the support of a team of engineers.¹⁶⁶ AI sandboxes are isolated testing environments where developers and researchers can experiment with AI and trial new models.

Despite the lack of high-quality data and challenges in access to infrastructure and computing power, civil society in Africa has been experimenting with building their own small models, trained on small datasets and for specific purposes rather than the general-purpose LLMs. Civil society is “leveraging AI for advocacy, factchecking, monitoring (of elections, hate speech and disinformation), digital literacy, public accountability, and community engagement”, according to CIPESA’s 2025 State of Internet Freedom Report, which assessed implications of AI on digital democracy in 14 African countries. ¹⁶⁷

“In the women’s sexual health and reproductive sector, a lot of the organisations have been able to draw patterns based on intake of complaints or issues to do with women’s SRHR [Sexual and Reproductive Health and Rights] issues. And they have been able to conduct research both, qualitative and quantitative, and some have built small models. You hear them saying we have built this app, we have built this tool to just see if it can teach women to do this, and when you explore some of these tools, one or another, you find that actually you see it is based out of data they have collected”, said Irene Mwendwa, a Nairobi-based consultant on technology, AI and law.She mentioned as an example, Zuzi, an “African, and trauma-informed” AI chatbot developed by South African organisation GRIT to provide information on the rights and legal services available to survivors of gender-based violence.¹⁶⁸

In other examples in Tunisia, independent media outlet Nawaat launched Nawaat Chronicles,¹⁶⁹ an AI‑enhanced archive platform allowing readers, researchers, students and journalists to easily access and search an archive of over 22 years of independent journalism on Tunisia in Arabic, French, and English. In Kenya, Mzalendo Watch has been developing AI tools to facilitate access to parliamentary data and proceedings.¹⁷⁰ The latter is one of several other examples of CSOs across the continent, including in Nigeria, South Africa, and Zimbabwe, developing AI tools to facilitate access to and analysis of official documents and improve transparency.¹⁷¹ In Senegal, Jangat, an AI-powered platform, allowed voters to compare the electoral programmes of candidates in the 2024 election.¹⁷²

“I have been able to see a lot of opt-in by citizens, opt-in from a point of civic duty or civic participation, where individuals are just using their knowledge on building technologies as computer scientists, as data scientists, some self-taught, some [are] young people who do not even have jobs, to really showcase how technology can be used to create social cohesion and civic engagement or public participation here in Kenya and other countries”, said Mwendwa. She added that it is unfortunate that participation and creativity are “inhibited” by government restrictions such as app restrictions and network shutdowns, as was the case during the Tanzanian election in 2025. “Africa can also harness the power of these same AI technologies to grow our democracies, our economic opportunities. At the same time, that same prowess is being used against them [African youth], communicating two very different things for young people who are just trying to seek opportunities”, she added.