South African study develops AI system for multilingual misinformation detection
A South African research project has developed an AI-based misinformation detection system capable of analysing content in English, isiZulu and Sepedi, addressing challenges faced by languages with limited digital resources.
A doctoral study in South Africa has developed and tested an AI system designed to identify misinformation across three languages: English, isiZulu and Sepedi. The research aims to address the limited availability of automated misinformation detection tools for African languages that lack large digital datasets.
The study was conducted by Dr Seani Rananga during her PhD at North-West University. Rananga, who is now a computer science lecturer at University of Pretoria, used COVID-19 misinformation as an initial case study to examine how AI models could support content verification in multilingual environments.
The research explored the use of multilingual AI models, machine translation and expanded training datasets to improve detection capabilities in languages with limited labelled data. The approach combined automated translation, human review and synthetic data generation to train and evaluate the system.
Results presented in 2025 showed that the performance of the models differed depending on the language and translation method used. Researchers noted that translation quality can directly affect detection accuracy, while cultural context, idiomatic expressions, sarcasm and code-switching remain difficult challenges for AI systems.
The system is currently a research prototype, but Rananga suggested that similar technologies could eventually support journalists, public institutions and communities during elections, public health crises and other situations where misleading information spreads quickly.
Future research is expected to examine misinformation shared through informal online conversations, including slang and mixed-language content, as well as misleading material in formats such as images, audio and video.
Why does it matter?
Many misinformation detection systems are designed primarily for English and other widely used languages with extensive digital resources. Developing AI tools for African languages such as isiZulu and Sepedi could improve access to verification technologies, but effective deployment will require locally developed datasets, native-language expertise and human oversight to account for cultural context and linguistic complexity.
