LLM-Based Localization in the Context of Low-Resource Languages

Recent advancements in large language models (LLMs) have enabled the development of systems capable of generating human-like responses across a wide range of tasks. However, research focus has been primarily on high-resource languages such as English, German, and French, whereas low-resource languages have not benefited from these advances. This has created a challenge for localization which requires multinationals to deploy natural language processing-based tools across world-wide geographic footprints. In this paper, we evaluate the state-of-the-art in question-answering for several low-resource Indian languages, including Hindi and Gujarati, and explore a sample Human Resources use case. We focus on neural machine translation based on transfer learning, multilingual meta-learning, and zero-shot approaches, combined with open source LLMs with conversational capabilities.