Gracebaytech/Yoruba-Knowledge-and-Cultural-Retrieval-

Development of a Yoruba Linguistic Knowledge Retrieval and Cultural Preservation System Using Retrieval-Augmented Generation (RAG). # Development of a Yoruba Linguistic Knowledge Retrieval and Cultural Preservation System Using Retrieval-Augmented Generation (RAG) ## Overview This repository contains the full implementation of a Yoruba Linguistic Knowledge Retrieval and Cultural Preservation System developed using Retrieval-Augmented Generation (RAG). The system is designed to support culturally grounded information retrieval and natural language generation for the Yoruba language, a low-resource African language with rich tonal, morphological, and cultural structures. The project integrates heterogeneous Yoruba textual resources and evaluates multiple retrieval strategies and RAG architectures across thematic domains including current affairs, religion, culture, entertainment, and social life. This repository accompanies the author’s postgraduate thesis submitted to the University of Ibadan. --- ## Research Objectives The objectives of this research are to: - To identify key linguistic features of Yorùbá language such as tone patterns, word structure, and cultural expressions through data analysis, and use these insights to build a more accurate and culturally aware RAG system. - To evaluate the impact of various retrieval techniques such as dense retrieval, sparse retrieval, hybrid approaches across the domains of current affairs, entertainment, religion, culture, and social life. - To optimise a Retrieval-Augmented Generation (RAG) system, including Naive, Advanced, and Modular variants, for enhancing text-based knowledge retrieval in the Yoruba language across the domains of current affairs, entertainment, religion, culture, and social life. - To investigate the influence of different large language models on the generation quality and cultural appropriateness of Yoruba text outputs across Naive, Advanced, and Modular RAG systems. - To assess the accuracy and relevance of RAG-generated outputs in Yoruba using quantitative metrics such as: MRR, precision and recall across different retr …