ENHANCING HEALTHCARE COMMUNICATION: A YORUBA-TO-ENGLISH NLP MODEL FOR DOCTOR-PATIENT INTERACTIONS USING MACHINE TRANSLATION AND SPEECH-TO-TEXT

Abstract

The problem of language barriers in healthcare is another problem in multilingual areas such as Nigeria where more than 500 languages are used. These obstacles inhibit effective communication between the medical staff and the patients, which results in misdiagnoses, poor adherence to treatment, and deteriorated outcomes. This study comes up with a Yoruba-to-English Natural Language Processing (NLP) model to handle this problem, and the study targets at improving the doctor-patient communication in the Yoruba-speaking areas. The model combines two major elements: Yoruba Speech Recognition (YSR) in order to turn Yoruba speech into text and Yoruba-English Machine Translation (YEMT) in order to translate the transcribed Yoruba text into fluent English. The model was trained by use of a transformer-based architecture on a dataset of Yoruba healthcare interactions, giving a Word Error Rate (WER) of 10.1% and a Sentence Accuracy of 77.0% on the test set. The findings prove that the model can enhance healthcare communication through minimizing the language barrier, and eventually patient care, diagnosis, and treatment compliance. Although informal speech and tonal differences may be a problem, the research can offer an encouraging way of a low-resource language in healthcare setting, which can trigger further studies and advancement in multilingual NLP systems. The study is part of the emerging body of research in NLP among African languages that provides a pragmatic framework which can be extended to other multilingual healthcare environment.