Multilingual Machine Translation Using Huggin...

Multilingual Machine Translation Using Hugging Face Models for AI-Powered Language Translation to Decode the World's Voices in Real-Time

Author Name : Dr. M. K. Jayanthi Kannan, Ritam Polley, Aditya Raj, Keshri Nandan, Dyutiman Bharadwaj, Parth Bindal

DOI: https://doi.org/10.56025/IJARESM.2024.1211242032

ABSTRACT A game-changing technology, machine translation (MT) makes it easy to communicate across linguistic barriers, which is essential for international trade, education, and cross-cultural interactions. This paper describes the creation of a multilingual translation pipeline that makes use of the mBART and NLLB models and Hugging Face's `transformers` library. While NLLB is tailored for low-resource languages, addressing issues such minimal linguistic data, the mBART model performs exceptionally well when translating high-resource languages. The pipeline guarantees precise and fluid translations with the use of sophisticated Transformer architecture and selfattention mechanisms. High translation quality was shown for languages such as English and Chinese in tests on a variety of datasets, while low-resource pairs showed adequate performance for contextual accuracy enhancement. To improve efficiency and scalability, the study highlights the value of preprocessing, fine-tuning, and performance measures like BLEU scores. This work underscores the potential of open-source NLP tools in enabling practical multilingual applications, contributing to accessible and effective cross-linguistic communication