Sanjita Phijam
2024
WMT24 System Description for the MultiIndic22MT Shared Task on Manipuri Language
Ningthoujam Justwant Singh
|
Kshetrimayum Boynao Singh
|
Ningthoujam Avichandra Singh
|
Sanjita Phijam
|
Thoudam Doren Singh
Proceedings of the Ninth Conference on Machine Translation
This paper presents a Transformer-based Neural Machine Translation (NMT) system developed by the Centre for Natural Language Processing and the Department of Computer Science and Engineering at the National Institute of Technology Silchar, India (NITS-CNLP) for the MultiIndic22MT 2024 Shared Task. The system focused on the English-Manipuri language pair for the WMT24 shared task. The proposed WMT system shows a BLEU score of 6.4, a chrF score of 28.6, and a chrF++ score of 26.6 on the public test set Indic-Conv dataset. Further, in the public test set Indic-Gen dataset, it achieved a BLEU score of 8.1, a chrF score of 32.1, and a chrF++ score of 29.4 on the English-to-Manipuri translation.