By Stephen Asante
Accra, Sept. 30, GNA – Scientists at the Kwame Nkrumah University of Science and Technology (KNUST) have developed ‘SignTalk-Gh’, an Artificial Intelligence (AI)-assisted model to support Ghanaian Sign Language recognition and translation in healthcare.
The innovation provides the first curated, domain-specific Ghanaian Sign Language dataset for healthcare, capturing doctor-patient conversations in healthcare settings.
The system uses advanced deep learning technologies to translate common healthcare terms and phrases, the Department of Computer Science said in a publication in the Scientific Reports Journal.
Other departments contributing to the work were Statistics and Actuarial Science, Disability and Rehabilitation Studies, Telecommunication Engineering, Meteorology and Climate Science, and Health Promotion and Disability Studies.
Dr Emmanuel Ahene, the Lead Researcher, said the dataset contained more than 9,000 video samples and accompanying metadata.
The metadata included sentence IDs, English translations, thematic categories, signer identifiers and temporal annotations.
“This vocabulary covers critical terminology such as anesthesia, cardiovascular, and symptoms, ensuring the dataset is robust for training AI models in complex healthcare scenarios,” Dr Ahene said.
He noted that the system was designed to address communication barriers faced by people with hearing and speech impairments, particularly in a country where domain-specific resources for critical sectors such as healthcare remained limited.
“By combining AI-assisted sentence generation, professional video recording, and rigorous annotation, the ‘SignTalk-Gh’ dataset serves as a powerful resource,” the Lead Researcher told the Ghana News Agency (GNA) in a briefing on the project.
He said the work involved close collaboration among healthcare providers, AI specialists and members of the deaf community to ensure that the dataset was culturally relevant, linguistically accurate and applicable to real-world healthcare environments.
“With a clear roadmap for model training, deployment, and future expansion, the SignTalk-Gh dataset underscores the potential of AI in driving inclusive healthcare, especially in underrepresented communities, and highlights the potential to address pressing societal challenges in communication,” Dr Ahene said.
The project’s problem statement indicated that despite the rapid advancement of AI technologies in natural language processing for spoken languages, sign languages, particularly those in Africa, remained vastly underrepresented.
It noted that the disparity was especially critical in healthcare, where effective communication could directly affect the quality and outcomes of care for deaf people.
“While AI-driven systems for sign language recognition have gained momentum globally, most of the existing models and datasets are centred around American or European sign languages, leaving African sign languages at the periphery of AI research,” the research team observed.
The researchers said the ‘SignTalk-Gh’ dataset could support the future development of recognition and translation systems, with demonstrated suitability for retrieval-based text-to-sign applications.
The project was carried out under the auspices of the Responsible Artificial Intelligence Lab and the Artificial Intelligence for Sustainable Development project.
It was funded by the French Embassy in Ghana and the Agence Française de Développement, with support from the UK Foreign, Commonwealth and Development Office, International Development Research Centre, and Artificial Intelligence for Development.
GNA
Edited by Agnes Boye-Doe
Reporter: Stephen Asante
[email protected]