Effective medical communication is crucial for diagnosis and patient safety, particularly in multilingual regions like North-West Nigeria. Hausa-speaking patients often face language barriers with English-speaking healthcare providers, resulting in poor health outcomes. Existing machine translation tools lack accuracy for low-resource languages like Hausa in medical contexts. This study uses a fine-tuned variant of the Helsinki-NLP OPUS-MT (en-ha) model using a curated 2,000 sized EMT dataset focused on medical terminology. The resulting transformer-based Neural Machine Translation model achieved strong performance with BLEU: 34.89, CHRF: 58.84, and ROUGE-L: 61.37. Notably, the fine-tuned model achieved a 68% improvement in BLEU score over the baseline OPUS-MT model (20.74).
Keywords: Neural Machine Translation, Medical Terminology, Transformer Model, Low Resource
Language
CISDI Journal Reference Format
Ladoja, K.T., Hamzat, Z.A. & Waziri, M. (2025): Domain-Specific Neural Translation: A Transformer-Based Model for Medical Terminology Explanation to Hausa Language. Computing, Information Systems, Development Informatics & Allied Research Journal. Vol 16 No 2, Pp 45-54. Available online at
isteams.net.
dx.doi.org