Empowering Marathi and Hindi Through LLMS: An Implementation of AI Applications in Translation, NLP, and STEM Localization

Abstract

The paper discusses the paradigm shift of enabling Marathi and Hindi with Large Language Models (LLMs) and focuses extensively on AI-based approaches to localization, translation, NLP, and STEM localization. The paper discusses the transformations of LLMs (GPT-3, BERT, IndicBERT) in local languages' online presence, particularly in India. AI and machine-learning techniques have fueled emphasis on Marathi-to-Hindi translations, and the refinement of STEM localization endeavors, like summarization. A host of challenges exist, including dialect differences and language intricacies, however, we see tremendous potential in AI-based technologies delivered bilingual English inclusion of Marathi and Hindi in life use and technical STEM uses, such as teacher support, community education, and STEM Education. The manuscript discusses advances in NLP tasks, such sentiment analysis and named-entity recognition (NER) in the local languages and engagement for users in local uses. Research explores how AI can support clients in sharing STEM content in local languages without barriers.