MedASL: The First Public Medical and Healthcare American Sign Language (ASL) Dataset

Several sign language translation datasets in the literature consist of videos, glosses, and texts. These datasets are designed for general-purpose daily communication, with only a few addressing communication in specific domains, such as weather, emergency, and public service. However, healthcare, a domain where precise communication is critical, remains underrepresented. The lack of medical sign language datasets limits the applicability of Sign Language Machine Translation (SLMT) systems in clinical settings, where seamless and accurate communication could improve individual care. Therefore, we introduce annotated datasets for healthcare. Consequently, we introduce MedASL, a continuous American Sign Language (ASL) corpus which focuses on medical communication, with gloss and text annotations. It is designed to support researchers and industry professionals in advancing SLMT systems. By incorporating medical terminologies and advanced data acquisition, such as the Intel RealSense camera, MedASL enables the development of accurate and context-aware models that reflect real-world healthcare scenarios. The dataset consists of 500 medical and healthcare-related statements, generated via prompt engineering using ChatGPT and signed by an ASL expert, simulating realistic dialogues between patients and healthcare professionals.