Generative Artificial Intelligence and Linguistic Epistemic Justice for isiZulu and African Languages in Multilingual Education
This article argues that achieving linguistic-epistemic justice for isiZulu and other African languages in AI-mediated education requires moving beyond mere text generation to establish a governance framework that treats these languages as valid infrastructures for reasoning, identity, and intellectual authority through equitable data representation, teacher agency, and institutional accountability.
Original paper licensed under CC BY 4.0 (https://creativecommons.org/licenses/by/4.0/). This is an AI-generated explanation of the paper below. It is not written or endorsed by the authors. For technical accuracy, refer to the original paper. Read full disclaimer
Imagine a classroom where a student asks a question in their home language, and a computer program answers back in the same tongue. For decades, the dream of artificial intelligence has been to speak every language fluently, acting as a universal translator that bridges gaps between people. But in the real world, these computer programs are not neutral tools; they are built on massive collections of text and data, much of which comes from English-speaking sources. This creates an uneven playing field. When a system is trained mostly on English, it becomes a master of that language but often stumbles when asked to reason, explain complex ideas, or tell a story in a language like isiZulu, one of South Africa's eleven official languages. The problem is not just about getting the words right; it is about whether the computer can understand the culture, the history, and the deep meaning behind those words. If a machine can only speak a language as a second-rate translation of English, it fails to respect the people who use that language as their primary way of thinking and learning.
Two researchers at the University of South Africa, Stella Bolanle Apata and Itumeleng I. Setlhodi, set out to investigate this exact tension. They did not run a new computer test to see if a specific chatbot could write a sentence in isiZulu. Instead, they gathered and studied forty-four different scholarly articles, policy documents, and technical reports to understand the bigger picture. They wanted to know if the rapid rise of generative artificial intelligence—the kind of technology that can write essays, solve problems, and hold conversations—was helping African languages grow or if it was quietly pushing them aside. Their work, a careful review of existing knowledge, suggests that the answer depends entirely on how schools and governments choose to use these tools. They found that without deliberate effort, these powerful systems will likely reinforce the dominance of English, treating African languages as mere channels for translating English ideas rather than as full, independent ways of building knowledge.
The researchers began by looking at the technical side of things. They found that while computers have gotten better at translating words, they still struggle with the deeper work of education. A model might produce a grammatically correct sentence in isiZulu, but it often lacks the cultural nuance or the specific vocabulary needed to explain a scientific concept or a historical event. This happens because the data used to teach these computers is scarce for African languages. The researchers describe this not as a failure of the languages themselves, which are rich and complex, but as a failure of the digital infrastructure. The computers simply have not been fed enough high-quality stories, textbooks, and conversations in isiZulu to learn how to think in that language. As a result, when students and teachers turn to these tools for help, they often find that the English version of the answer is clearer, more detailed, and more reliable. This creates a subtle pressure to switch to English, not because the user prefers it, but because the technology makes it the path of least resistance.
The study goes deeper than just technical glitches to examine the human and political side of the equation. The authors argue that language is more than a tool for communication; it is a vehicle for identity and authority. When a computer system treats an African language as a secondary option, it sends a message that the knowledge and culture associated with that language are less valuable. This is what the researchers call "epistemic injustice," a fancy term for the unfair treatment of people as thinkers and knowers. If a student cannot use their home language to argue a point, solve a math problem, or write a research paper with the same confidence as an English speaker, they are being denied full participation in the world of ideas. The researchers warn that if schools simply adopt these tools without a plan, they risk automating this inequality, making it harder for African languages to survive and thrive in the digital age.
However, the paper does not suggest that artificial intelligence is inherently bad for African languages. The authors see a path forward, but it requires a shift in how we think about these tools. They propose a new way of governing these systems, one that puts language equity at the center. This means that before a school or university buys a new AI tool, they should ask specific questions: Does this tool support our local languages well? Has it been tested by people who speak those languages? Are the people who created the data for this tool being recognized and rewarded? The researchers suggest that teachers play a crucial role here. They need to be trained not just on how to use the technology, but on how to spot when the computer is getting it wrong. A teacher must be able to look at an AI-generated answer in isiZulu and decide if it captures the true meaning of a lesson or if it is just a shallow imitation.
The researchers also emphasize that the responsibility lies with institutions, not just individuals. Universities and governments cannot simply say they support multilingualism and then leave the details to chance. They need to create policies that require AI companies to be transparent about their data and to invest in building better resources for African languages. This includes creating digital libraries of local knowledge, developing better ways to test these systems, and ensuring that the people who speak these languages are involved in the design process. The goal is to move from a situation where African languages are just an afterthought to one where they are treated as essential parts of the digital world.
In the end, the paper reframes the question from "Can the computer speak isiZulu?" to "Does the computer respect isiZulu?" The answer, the authors suggest, is not a simple yes or no. It is a conditional possibility. The technology has the potential to support and strengthen African languages, but only if we actively build the infrastructure, train the teachers, and write the policies to make it happen. Without these steps, the future of education in multilingual societies like South Africa could become a place where English remains the only language of serious thought, while other languages are left behind. With these steps, however, artificial intelligence could become a powerful ally in preserving cultural memory and expanding access to knowledge for everyone. The work of Apata and Setlhodi serves as a clear roadmap for this journey, urging us to look beyond the flash of new technology and focus on the quiet, essential work of justice and inclusion.
Drowning in papers in your field?
Get daily digests of the most novel papers matching your research keywords — with technical summaries, in your language.