Browsing by Author "Olawuyi H. O."

Now showing 1 - 1 of 1
  • No Thumbnail Available
    Item
    Human-Computer Interaction-Based Yorùbá Language using Recurrent Neural Network of Deep Learning
    (Faculty of Computing and Information Technology and Faculty of Natural and Applied Science, Sule LAmido University, Kafin Hausa, 2025-03) Babatunde A. N.; Ajao J. F.; Babatunde R. S.; Olawuyi H. O.; Yusuff, S. R.
    Background: This paper presents the development of a Human-Computer Interaction-Based Yorùbá to English Language Translator using a Recurrent Neural Network (RNN), specifically a Bidirectional Gated Recurrent Unit (GRU) encoder-decoder model. This study presents a novel contribution to existing studies by systematically addressing the setbacks of Yoruba tonal variations with larger corpus and offering a user -friendly interface both underexplored in prior research. Aim: The research aims to develop a machine translation system for translating Yoruba to English and English back to Yoruba applying Bidirectional Gated Recurrent Unit (GRU) algorithm. Methodology: 10054 instances of bilingual Yorùbá and English parallel corpus were collected from Kaggle. The dataset was split into 80% and 20% for training and testing respectively. The Bidirectional GRU model was trained to learn the relationships between the languages with optimization techniques such as gradient descent and backward propagation. Results: The model was evaluated using the BLEU score achieving 93% and 96% accuracy on the training and test sets respectively indicating its robustness in translating between Yorùbá and English and vice versa. A user-friendly web interface was integrated for easy real-time translation. The final system was deployed on a cloud server for remote access. The findings demonstrates that models lik e the Bidirectional GRU can ef fectively handle Yorùbá’s linguistic complexities, improving translation accuracy and facilitating cross-cultural communication. The system implemented in this paper performs well with standardized text but its performance may reduce for high idiomatic expressions. Notwithstanding, the system holds promise for applications in education, multilingual content localization, and assistive communication tools, particularly for Yorùbá-speaking communities.