DEVELOPMENT OF NEURAL YORUBA-TO-ENGLISH MACHINE TRANSLATION SYSTEM USING NLP TECHNIQUES

Abstract This study presents the development of a Neural Yoruba-to-English Machine Translation (MT) system using Natural Language Processing (NLP) techniques. Yoruba, a low-resource African language with tonal complexity and rich idiomatic expressions, poses unique challenges for computational processing. To address these, a curated Yoruba-English parallel corpus was collected and preprocessed through text cleaning, normalization, tokenization, and alignment. Two neural architectures were implemented: a sequence-to-sequence model with attention and a Transformer model. The models were trained and evaluated using BLEU scores and human assessments by bilingual Yoruba-English speakers. Results revealed that the Transformer model achieved superior performance, with a BLEU score of 32.5 compared to 23.8 for the Seq2Seq model. Human evaluations also favored the Transformer for fluency, adequacy, and cultural appropriateness, highlighting its effectiveness in handling tonal distinctions and idiomatic expressions. The findings demonstrate that Transformer-based architectures, combined with careful preprocessing, significantly improve translation quality for low-resource languages. This research contributes to advancing African language processing, promoting digital inclusion, and preserving Yoruba linguistic heritage while providing practical applications in education, commerce, and communication.