A recent study conducted as part of a doctoral dissertation has led to the creation of multilingual artificial intelligence (AI) tools capable of detecting false information in South African languages. This solution aims to address the underrepresentation of African languages in the field of AI.
Scientific Achievements and Research Focus
Dr. Siani Rananga, a lecturer in the Department of Computer Science at the University of Pretoria, demonstrated that multilingual AI can identify misleading information in IsiZulu, English, and Sepedi. Her dissertation, completed at Northwestern University, focused on areas such as natural language processing, multilingual AI, and disinformation detection in African languages, gaining international recognition.
As a pilot case for testing the multilingual structure, Rananga used disinformation related to the Covid-19 pandemic. She emphasized that this research was motivated by the need to ensure that speakers of indigenous South African languages are not left behind as the significance of AI in identifying fake news grows.
Challenges of Linguistic Diversity
Rananga noted that South Africa is one of the most linguistically diverse countries in the world, yet most AI systems for disinformation detection are developed primarily for English. This puts native speakers of local languages at a disadvantage when seeking reliable information online.
In her dissertation, Rananga initially focused on studying English, isiZulu, and Sepedi separately to build a solid foundation. The research covered both machine translation models and multilingual AI models. Machine translation helped create linguistic resources where data was scarce, while multilingual models were trained to understand isiZulu and Sepedi directly.
Challenges in Resource Creation
Native speakers participated in the project to verify the quality of translations. Nevertheless, preserving cultural meaning and idiomatic expressions remains an ongoing challenge, requiring further collaboration with linguists and native speakers. Rananga pointed out that the most difficult task was not creating the AI model itself, but forming high-quality linguistic resources.
Given the limited labeled disinformation data for African languages, she combined machine translation, human verification, and synthetic data generation to develop robust training materials for isiZulu and Sepedi. One key finding of her work was that investment in high-quality African language resources is just as important as developing more sophisticated AI models.
Prospects for System Implementation
Describing a plan to implement this structure to protect communities during critical events such as elections or public health crises, Rananga clarified that the current structure is a research prototype, but the long-term vision is much broader. She sees it as part of a multilingual fact-checking platform where users can verify suspicious messages, images, audio, or videos before they are disseminated.
In the event of elections or health emergencies, such a system could support fact-checkers, journalists, government, and the public by providing quick, evidence-based checks in South African languages. Rananga also noted that one of the hardest tasks for AI is that lies often rely on sarcasm, irony, or local humor, rather than obvious false facts. Understanding sarcasm, humor, and cultural context requires analyzing not only words but also intent and context.
Future Research Directions
Rananga added that although AI is improving, it remains an active area of research. Her future work will focus on incorporating conversational context, multimodal information such as images and videos, and explainable AI to enhance the transparency and reliability of these systems. Another challenge she highlighted is that South Africans often switch between languages mid-sentence. The next phase of research will focus on training and evaluating the structure on real social media conversations that include code-switching, slang, and local expressions to better reflect the natural communication of South Africans online.
Rananga stressed that the development of AI for African languages requires collaboration. She expressed hope to make models, benchmarks, code, and other research resources available wherever possible, while adhering to ethical, legal requirements, and data sharing mandates. The goal is to support other researchers and accelerate the development of AI technologies that better serve African languages and communities.
Dr. Rananga received the Google PhD Fellowship, one of the leading global awards supporting doctoral research in AI. She also received the 'Best Poster' award at Deep Learning IndabaX South Africa in 2025, which provided her funding to present her research at the Pan-American University in Lagos, Nigeria, in August 2026.
Professor Sihavu Ngubane, a culture expert from the University of KwaZulu-Natal, congratulated Rananga on this achievement. He stated that the research will play an important role in the development and promotion of indigenous languages, helping them reach the global stage, as South African policy allows speaking all 12 official languages, and her intention to include them all will be significant in this country, which boasts immense cultural and linguistic diversity.
