Urdu to English Neural Machine Translation using Transformers

Abstract

Neural machine translation is a field of research that has gained worldwide interest over the years. It involves the use of deep learning algorithms to translate texts from one language to another. The research in this field has significantly advanced over the years, with numerous state-of-the-art techniques being developed to improve the accuracy and speed of the translation process. Despite the considerable progress made in neural ma chine translation, the development of Urdu to English translation systems is still lacking. This can be attributed to the complex nature of the Urdu language, which is charac terized by a unique writing system and complex morphology. Additionally, the lack of standardized datasets and linguistic resources for Urdu poses a significant challenge in the development of Urdu to English translation models. In this research, we present a neural machine translation system specifically designed for translating text from Urdu to English. Our primary focus is to improve the accuracy of previous translation models by employing a transformer-based approach. our developed system achieves a commend able BLEU score of 45.58, WER Score of 0.41 indicating competitive performance in the domain of Urdu-to-English translation. This noteworthy result further establishes the efficacy of our approach and highlights the potential for its application in practical translation tasks.