MedQA-MA: A Moroccan Arabic medical question-answering dataset for virtual healthcare assistants and large language models

The healthcare domain constitutes a fundamental pillar of national development, as maintaining population health not only enhances citizens' quality of life but also generates substantial economic benefits through increased productivity, innovation, and workforce participation. However, the healthcare industry faces numerous challenges and barriers that impede universal access to medical services. In low- and middle-income countries, significant portions of the population forego medical consultations due to various socioeconomic constraints, including prohibitive consultation fees, scheduling difficulties, and extended waiting periods. Consequently, there is an urgent need for innovative approaches to optimize healthcare delivery processes. Recent advances in artificial intelligence have demonstrated promising potential in developing intelligent systems that address healthcare accessibility gaps. These innovations include medical chatbots, appointment booking systems, disease-prediction models, and psychiatric virtual assistants. However, such technological enhancements have predominantly focused on high-resource languages, while research in low-resource languages, particularly Arabic, remains in its preliminary stages. This disparity is especially pronounced in Arabic dialects, which differ substantially from Modern Standard Arabic in terms of vocabulary, syntax, and semantic structures. To address this critical gap, we present the first comprehensive dataset for the Moroccan Arabic dialect in the healthcare domain. The MedQA-MA dataset comprises 108,943 question-answer pairs in text format, with each pair categorized according to medical specialty. Including 23 distinct medical specialties, this dataset serves multiple applications, including sentiment analysis, specialty classification, question-answering systems, and the development of human-like medical chatbots. The dataset has been meticulously curated, annotated, and validated by qualified medical professionals, ensuring its reliability and clinical relevance for developing realistic healthcare systems grounded in authentic medical interactions. The MedQA-MA dataset is publicly available and freely accessible at https://data.mendeley.com/datasets/v6gs7nsy9z/1, representing a significant contribution to Arabic Natural Language Processing research in healthcare applications and facilitating the development of culturally and linguistically appropriate medical AI systems for Arabic-speaking populations.