Language Translation Using Marine MT

Language Translation Using Marine MT

Adithya V CEI(AIML)

Presidency University Bengaluru,Karnataka,India ADITHYA.20211CEI0057@presidenc yuniversity.in

Tejas Gowda V CEI(AIML)

Presidency University Bengaluru,Karnataka,India TEJAS.20211CEI0051@presidencyuni versity.in

Poornima Selvaraj

Senior Associate Proffesor Presidency University Bengaluru,Karnataka,India

poornima.s@presidencyuniversity.in

Abstract—In today’s global landscape, effective multilingual communication is essential. Traditional machine translation methods, including rule-based and statistical approaches, have been surpassed by neural machine translation (NMT), particularly with the adoption of transformer-based models. Hugging Face, a widely-used open- source platform, offers pre-trained NMT models such as MarianMT, M2M-100, and No Language Left Behind (NLLB), enabling efficient and scalable translation across numerous language pairs.

This paper explores the implementation of Hugging Face’s translation models in real-time multilingual applications, including live chat, conferencing, and dynamic content translation. We evaluate the platform’s ease of deployment, model versatility, and performance in latency-sensitive environments. Additionally, challenges such as low-resource language support, context preservation, and real-time processing constraints are examined.

Our findings highlight Hugging Face’s potential in accelerating research and development in NMT, offering practical solutions for seamless cross-language communication in real-world systems.

Keywords—Language     translation,     Hugging     Face, Python, styling, Microsoft, Textmining