Scholarlink

The process of automatically converting the text from one language to another natural language is Machine Translation. Urdu is a language that is spoken in Pakistan and in some part of India, Afghanistan, and Middle East. Only 5 to 10% people can understand English language. They are not able to get latest knowledge in this digital globe. That’s why, this study explores the development of English to Urdu and Urdu to English machine translation system using transfer learning-based approaches. The research aims to address the language barrier that limits access to information for Urdu speakers by leveraging the FB mBART-50 model, a state-of-the-art multilingual sequence-to-sequence model. Through extensive evaluation, the FB mBART-50 model demonstrated significant improvements over previous models, achieving maximum BLEU score from the combined data is approximately 0.867 for English to Urdu translation and 0.669 BLEU score for Urdu to English translation. Despite some challenges with pronoun accuracy and lexical repetition, the model consistently produced contextually appropriate and semantically accurate translations. This study highlights the potential of deep/transfer learning methods to enhance machine translation quality, suggesting that with further improvements and optimizations, these methods can facilitate seamless communication in multilingual environments.

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

Enhancing English–Urdu Machine Translation Using Transfer Learning

  • 会议论文

  • 发表日期：

    2026年

查看全文
  • Fida Ullah,
  • M. Tayyab Zamir,
  • M. Ahmad,
  • Irina Gelbukh,
  • Grigori Sidorov,
  • Edgardo Manuel Felipe Riverón

摘要

The process of automatically converting the text from one language to another natural language is Machine Translation. Urdu is a language that is spoken in Pakistan and in some part of India, Afghanistan, and Middle East. Only 5 to 10% people can understand English language. They are not able to get latest knowledge in this digital globe. That’s why, this study explores the development of English to Urdu and Urdu to English machine translation system using transfer learning-based approaches. The research aims to address the language barrier that limits access to information for Urdu speakers by leveraging the FB mBART-50 model, a state-of-the-art multilingual sequence-to-sequence model. Through extensive evaluation, the FB mBART-50 model demonstrated significant improvements over previous models, achieving maximum BLEU score from the combined data is approximately 0.867 for English to Urdu translation and 0.669 BLEU score for Urdu to English translation. Despite some challenges with pronoun accuracy and lexical repetition, the model consistently produced contextually appropriate and semantically accurate translations. This study highlights the potential of deep/transfer learning methods to enhance machine translation quality, suggesting that with further improvements and optimizations, these methods can facilitate seamless communication in multilingual environments.