Lost in translation no more

Researchers focusing on zoonotic disease and computer science may help shape a future where life-saving health knowledge truly speaks everyone’s language.

When a child is bitten by a dog in a rural village, the difference between life and death can hinge on something deceptively simple: access to clear, understandable information.

While rabies is a preventable disease, it still claims lives in South Africa, often because people do not receive timely or accurate guidance. One reason is that most public health information about rabies is available only in English, effectively excluding many communities for whom English is not a first language.

New research by the University of Pretoria (UP) is tackling this problem by asking a powerful question: what if artificial intelligence (AI) could speak to people about health risks in their own language?

The study brought together zoonotic disease expertise from the national Department of Health’s National Institute for Communicable Diseases and UP’s Future Africa platform, as well as AI research expertise from the University’s Department of Computer Science.

Using rabies as a pilot case, researchers focused on Sepedi, one of South Africa’s 11 official languages and a “low-resource” language in the digital world, meaning there is limited online data available to train AI systems.

They translated and carefully validated rabies awareness knowledge in Sepedi, ensuring it was not only linguistically correct but also culturally appropriate and factually accurate. This content was then used to fine-tune two publicly accessible large language models: OpenAI’s GPT-4o and Google’s Gemini 1.5 Flash.

“Rather than building an AI system from scratch, the researchers took an innovative, practical approach,” Dr Tedson Nkoana of Future Africa says. “They adapted existing, widely used models and tested how well these could handle health information in a local African language.”

The results were encouraging. Across measures such as fluency, clarity, accuracy and cultural relevance, the AI performed well enough to support a prototype chatbot that can answer rabies-related questions in Sepedi.

According to Prof Wanda Markotter, who holds the Chair in One Health at Future Africa, the potential benefits to society are significant.

“A language-inclusive chatbot could provide real-time educational support in rural clinics, help community health workers generate awareness materials in multiple languages, and give the public direct access to reliable zoonotic disease information via mobile phones or the web. In the long term, tools like this could help reduce rabies transmission and fatalities among both people and domestic animals, while also promoting trust in health messaging.”

Equally important is what this research represents beyond rabies, says Prof Vukosi Marivate, the ABSA Chair of Data Science.

“It shows how AI can be used to advance linguistic equity, rather than deepen digital divides.”