DESIGN AND IMPLEMENTATION OF AN ENGLISH–YORUBA NEURAL MACHINE TRANSLATION SYSTEM USING WORD2VEC

ABSTRACT 53 This study developed a bidirectional English Yoruba text-to-text translation system using Python, with the objective of enhancing contextual accuracy and promoting the preservation of the Yoruba language in digital communication. The research addressed the limitations of existing translation tools, which often struggle with the cultural depth, tonal complexity, and semantic nuances of Yoruba. The methodology integrated Word2Vec for semantic word embeddings and tokenization techniques to process and analyse both English and Yoruba texts. Word2Vec maps words into a multidimensional vector space to capture semantic relationships, while tokenization ensures accurate handling of linguistic structures and tonal variations unique to Yoruba. These techniques are combined to produce translations that go beyond literal word for word mapping, enabling deeper contextual understanding and more culturally faithful outputs. The theoretical framework was grounded in computational linguistics and natural language processing, emphasizing the importance of semantic modelling in bilingual translation systems. The system was deployed through a user-friendly interface to ensure accessibility for a wide range of users, including Page 13 of 79 - Integrity Submission vi Submission ID trn:oid:::2945:316981830 Page 14 of 79 - Integrity Submission 21 Submission ID trn:oid:::2945:316981830 educators, linguists, and native speakers. Improved bidirectional translation accuracy, effective resolution of semantic ambiguities, and preservation of cultural references were found from the evaluation, demonstrating superior performance compared to conventional translation tools. Based on these findings, the study affirmed that integrating machine learning techniques with linguistic insights can yield scalable, context-aware translation solutions that promote language preservation and digital inclusivity. Future work should focus on expanding the training dataset with diverse linguistic sources, incorporating advanced neural models such as Transformers, and deploying the system as a web-based or mobile application to support real-time translation, increased accessibility, and broader adoption across various user demographics. Keywords: English–Yoruba Translation, Word2Vec, Semantic Modelling, Tokenization, Language Preservation