Neural Machine Translation for Hindi to English Using Sequence-to-Sequence Architecture with Attention Mechanism
Ishan Rathi, Paurav Goel, Jaspreet Singh · 2024
In an increasingly linked world, effective cross-linguistic communication has become essential. Machine translation systems play an important role in improving communication among speakers of different languages. This research paper describes a thorough analysis into the creation and testing of a machine translation framework for translating Hindi words into English. The model is trained using a sequence-to-sequence (Seq2Seq) architecture and an attention mechanism on a dataset of concurrent Hindi-English phrase pairings. The training phase, which lasts 50 epochs, produces encouraging results, with the model reaching a training accuracy of 81.68% along with a validation accuracy of 75.71% on a rigorously selected dataset of different English-Hindi phrase pairings. The qualitative examination of the translated result reveals great accuracy and integrity to the original purpose. Comparative comparison using a benchmark translation model confirms the proposed model's ability to capture language intricacies and provide contextually suitable translations. This paper also identifies limits and suggests areas for further research, including the use of advanced approaches like domain adaptation along with transfer learning to improve translation quality. Overall, this study advances machine translation technologies and has implications for increasing cross-lingual communication in many linguistic circumstances.