RealTalk
Loading...
Files
Date
2025
Journal Title
Journal ISSN
Volume Title
Publisher
NHCE
Abstract
Language is one of the most fundamental ways humans communicate, but linguistic diversity often poses challenges in understanding and accessibility. This project aims to bridge these gaps by developing a robust multilingual translation system powered by state-of-the-art Marian MT models from Hugging Face's transformers library. The system supports translations from English into 13 diverse languages: Hindi, Tamil, Malayalam, Kannada, Telugu, Gujarati, Bengali, French, German, Russian, Japanese, Chinese, and Lebanese. By combining pretrained machine translation models with an efficient and scalable implementation, the system ensures translations are not only linguistically accurate but also contextually relevant, capturing cultural nuances wherever possible. The modular design enables straightforward integration of additional languages, while its flexibility accommodates varying linguistic complexities and syntactical structures. This capability positions the system as a versatile tool for education, healthcare, business, and cultural preservation.
A major innovation of this project lies in its ability to address the limitations of pretrained models for underrepresented languages like Lebanese by setting a foundation for custom model training. The project explores solutions to technical challenges such as memory optimization, model compatibility, and accurate handling of idiomatic expressions. The inclusion of future-facing goals, such as real-time translation and enhanced contextual understanding, underscores its scalability and relevance in an increasingly globalized world.