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.
