A Bidirectional Neural Machine Transliteration System for Manipuri Bengali and Latin Scripts

Kshetrimayum Boynao Singh, Asif Ekbal, Partha Pakray

Abstract: Transliteration plays a crucial role in the bridge between linguistic and scriptural diversity, enabling seamless communication across different languages and writing systems. In regions like Manipur, where the Manipuri language is written in multiple scripts, such as Meitei, Bengali, and Latin, the need for effective transliteration tools becomes paramount. In this paper, we propose the development of a bidirectional neural machine transliteration system to convert text between Manipuri Bengali and Manipuri Latin scripts. Leveraging the Transformer architecture implemented in the Neural Machine Translation framework, our system achieves accurate and fluent transliteration in both forward and reverse directions. We present detailed results of experimentation and evaluation in a parallel corpus of Manipuri text, demonstrating the effectiveness of our approach in capturing complex phonetic mappings and achieving high transliteration accuracy.

External IDs:doi:10.36227/techrxiv.176739468.82264211/v1