L-Semakale/sesotho-medical-mt

# Sesotho Medical Machine Translation System ## Project Overview The **Sesotho Medical Machine Translation System** is a proof-of-concept bilingual medical translation web application designed to support healthcare-related translation between **English and Sesotho**. Healthcare communication in Lesotho is often affected by a language barrier. Many patients speak Sesotho as their main everyday language, while clinical records, medication labels, treatment instructions, referral notes, discharge instructions, consent information, and health guidelines are commonly written in English. This project addresses that gap by providing a specialized medical translation prototype for healthcare communication and research. The system uses a **three-layer hybrid translation pipeline** to improve transparency, reduce reliance on unconstrained neural generation, and support safer translation in the medical domain: 1. **Exact Corpus Match** 2. **Semantic Similarity Search** 3. **Fine-tuned Neural Machine Translation using NLLB-200** A medical safety filter is included to flag high-risk terms and provide warnings where professional clinical review is required. > **Corpus Note:** The project uses a 5,000-pair curated English–Sesotho medical corpus. Of these pairs, 4,536 were reviewed, 459 remained raw, and 5 reached full clinical verification status. The corpus is therefore suitable for proof-of-concept research but should not be treated as a fully clinically verified medical translation resource. > **Medical Disclaimer:** This system is a proof-of-concept research prototype. It is not a certified medical device and must not be used as the sole basis for diagnosis, treatment, medication dosing, emergency care, or clinical decision-making. All translations should be reviewed by a qualified bilingual healthcare professional before clinical use. --- ## Key Features - English-to-Sesotho and Sesotho-to-English interface support - Three-layer hybrid translation pipeline - Curate …