BHASHA SETU: A MODULAR MULTILINGUAL TRANSLATION SYSTEM FOR LOW RESOURCE AND MULTISCRIPT ENVIRONMENTS

Abstract

Translation systems currently in use often fall woefully short in environments with spotty connectivity and diverse linguistic requirements largely because offline functionality and image-based content are frequently unsupported alongside low-resource languages spoken in India. Essential is a modular multilingual translation solution guaranteeing privacy and adaptability across many languages in various formats quite seamlessly. This work presents an amazing multilingual document translating system with image-based translating capability as well as offline and online translating capability. It mostly depends on neural machine translation to enable several languages smoothly and makes good use of optical character recognition (OCR) to extract text from images. The system uses external APIs to prioritize privacy by providing offline translating between English and Marathi without an internet connection, so enabling real-time translation. Its modular architecture guarantees effective translation paths and general compatibility with many formats, including PDF and picture files. These days, it readily and effectively negotiates language obstacles in many different fields. The paper closes with evaluating performance and making recommendations for future improvements including progressively using more sophisticated NLP models and including more eccentric language support.