MedQA-Darija-MultiLingual

MedQA-Darija-MultiLingual The largest open trilingual medical Q&A dataset with directly-playable speech audio for English, French, and Moroccan Darija.** A research dataset for the BRAIN HEALTH initiative, designed for multilingual medical NLP, low-resource speech recognition, healthcare chatbots, and clinical education tools targeting Morocco and the broader Maghreb region. Dataset is currently in **scientific validation phase**. After programmatic validation (Stage 1 LOF outlier detection, Stage 2 Causal-DAG consistency, Stage 3 linguistic & medical NER coverage), it will be reviewed by licensed medical professionals before peer-reviewed publication. At a glance | Property | Value | |---|---| | Total Q&A pairs | **267,627** | | Specialties covered | **71** | | Languages | English · French · Moroccan Darija (Arabic script) | | Audio modality | 6 MP3 per row (Q+A × 3 languages) | | Real-world data | 163,488 pairs (open medical corpus + iCliniq scraped) | | Synthetic data | 104,139 pairs (multi-LLM generated, validated) | | Audio files | ~427k MP3 (~91 GB) on Hugging Face Xet storage | | Total file size | ~94 GB | Configurations `default` — Synthetic (104,139 rows) Trilingual synthetic Q&A generated by multiple LLM providers (Mistral, Cerebras, Groq, GPT-4o-mini, OpenRouter). Every row carries six audio cells (Q+A in EN/FR/Darija), playable directly in the Hugging Face Viewer. `real` — Real-world scraped (163,488 rows) Real medical Q&A scraped from the open medical web. Two sub-sources: **scraped_icliniq** (3,383 rows) — patient questions and licensed-doctor answers from publicly available iCliniq.com pages, manually trilingualized and audio-rendered. **open_medical_corpus** (160,105 rows) — open medical corpora translated trilingually. iCliniq rows appear first in the `real` split; both sub-sources share identical schema with audio columns (audio cells are populated where speech files are available, else `null`). Schema (per row) A single row contains the same medic