Advancements in Geospatial Question-Answering Systems: A Case Study on the Implementation in the Kazakh Language
Alibek Barlybayev, Assel S. Mukanova · 2024
This research delineates a comprehensive framework for the development of a Question-Answering System tailored to the Kazakh language, underscoring its significance within the domain of Text Processing for Low Resource Languages. The initiative aims to mitigate the deficiency in digital tools for underrepresented languages, with a particular emphasis on geographical inquiries pertaining to Kazakhstan. This endeavor enhances both the accessibility and comprehension of the nation's geographical data. Addressing the inherent challenges of Low Resource Languages text processing, the methodology encompasses the formulation of a specialized question-answer corpus, the employment of a Bidirectional Encoder Representations from Transformers-based model, and the application of the Bilingual Evaluation Understudy metrics for system evaluation. The research commences with the meticulous assembly of a corpus consisting of 50,000 questions, setting the foundation for subsequent developmental stages and ensuring the robustness of the Question-Answering System. In the ensuing phase, a Transformers model with 91,821,056 parameters is meticulously trained to adeptly capture the linguistic intricacies of the Kazakh language. The culmination of this project involves an exhaustive evaluation utilizing Bilingual Evaluation Understudy metrics, where the system garners a commendable average score of 0.9576, indicating a substantial concordance between the machine-generated responses and the reference answers. This metric substantiates the system's efficacy in decoding and addressing queries related to Kazakh geography. The outcomes of this study contribute significantly to the field by delineating a systematic and in-depth approach to Question-Answering System development and exemplifying the model's effectiveness through detailed evaluation and comparative analysis.