A Systematic Review on Hindi to English Machine Translation Using Machine Learning

Effective cross-language communication is vital for fostering collaboration between countries. This paper reviews various machine translation (MT) techniques, focusing on the challenging task of translating Hindi to English. Techniques covered include “Rule-Based Machine Translation (RBMT)”, “Statistical Machine Translation (SMT)”, “Example-Based Machine Translation (EBMT)”, and “Hybrid Machine Translation”, each with distinct advantages and limitations. The literature review highlights hybrid models combining RBMT, SMT, and EBMT for enhanced translation quality, as well as NMT systems leveraging linguistic features. A case study on a Transformer-based NMT system, trained with the Opennmt-tf toolkit on NVIDIA Tesla K80 GPUs, examines word and subword-level tokenization and incorporates back-translated data to enhance performance. Results show that subword-level tokenization outperforms word-level tokenization, achieving a peak BLEU score of around 22, compared to approximately 21 for word-level tokenization. These findings underscore the benefits of subword tokenization and back-translation in improving translation accuracy. The systematic review, following PRISMA guidelines, includes 15 selected research publications. This study emphasizes the need for ongoing innovation in Hindito-English MT systems to improve accuracy and support crosslanguage communication.