article · Scientific Reports
African sign languages, including Ghanaian Sign Language, remain underrepresented in artificial intelligence tools due to their multimodal nature involving hand gestures, facial expressions, and spatial movements. The lack of standardisation, regional variations, and an absence of large public datasets have limited the development of translation and recognition systems. This gap creates significant communication barriers for individuals with hearing impairment in critical settings such as healthcare. To address this, a new resource named SignTalk-Gh has been introduced. It is the first curated, domain-specific Ghanaian Sign Language dataset focused on healthcare, capturing doctor-patient interactions to support translation research and retrieval-based text-to-sign synthesis. The work establishes methodology for data collection, annotation, validation, and evaluation, aiming to enable future neural machine translation models for medical environments and eventually other sectors.
Effective communication between healthcare providers and patients is vital for safe and accurate medical treatment. By addressing the severe shortage of digital resources for Ghanaian Sign Language, this work provides the foundational data needed to build assistive technologies. This helps bridge communication gaps for deaf individuals, fostering better healthcare access and greater social inclusion across underrepresented linguistic communities.
This dataset supports the creation of AI-driven recognition, translation, and retrieval-based text-to-sign synthesis systems for healthcare providers and deaf patients. The technology appears to be at an early stage of development, as the resource is intended to enable the future building and training of neural machine translation models rather than functioning as a market-ready clinical product.
AI-generated from the published abstract. Always read the original work before citing.
Despite rapid advancements in AI-driven language processing tools, sign languages across Africa, including Ghanaian Sign Language (GhSL), remain critically underrepresented. Unlike spoken languages, sign languages involve complex hand gestures, facial expressions, and spatial movements, making their computational processing inherently multimodal. Challenges such as the lack of standardization, regional variations, and the absence of large, publicly available datasets have hindered the development of AI-powered sign language recognition and translation systems, particularly in underrepresented communities, leading to lack of accessibility and inclusion for persons with hearing impairment, especially in essential areas like healthcare, where effective communication is vital. Currently, no publicly available Ghanaian Sign Language datasets on healthcare access and interactions exist. This highlights the need for a comprehensive, standardized dataset to support robust research and practical applications. To address this, we introduce SignTalk-Gh, the first curated domain-specific Ghanaian Sign Language dataset for healthcare, designed to support future development of recognition and translation systems, with demonstrated suitability for retrieval-based text-to-sign applications. This dataset is specifically designed to aid translation research, capturing doctor-patient conversations in Ghanaian healthcare settings. This paper outlines the methodology behind the dataset’s construction, including data collection, annotation, validation, and evaluation processes. We also discuss its potential applications in AI-driven healthcare accessibility and its role in advancing research on African sign languages. Future directions include expanding the dataset to other domains, such as education, general conversations, and, among others, subsequently developing neural machine translation models for GhSL interpretation in medical environments and other domains
This page summarises published work. The authoritative version sits with the publisher.
DOI: 10.1038/s41598-026-43478-9
Is something wrong with this record? Report it or request removal.
Discussion
Have you built on this work, tried to replicate it, or seen it applied in practice? Share what you know. Verified researchers and MARATTO™ domain experts can open a discussion, and any member can reply. Contributions are reviewed before they appear.
No discussion yet. Open the first thread.
New to MARATTO™? Create a free account.
