SH at WMT25 General Machine Translation Task

Hayate Shiroma


Abstract
I participated in the unconstrained track of the English-to-Japanese translation task at the WMT 2025 General Machine Translation Task.My submission leverages several large language models, all of which are trained with supervised fine-tuning, and some further optimized via preference learning.To enhance translation quality, I introduce an automatic post-editing model and perform automatic post-editing.In addition, I select the best translation from multiple candidates using Minimum Bayes Risk (MBR) decoding.For MBR decoding, I use COMET-22 and LaBSE-based cosine similarity as evaluation metrics.
Anthology ID:
2025.wmt-1.39
Volume:
Proceedings of the Tenth Conference on Machine Translation
Month:
November
Year:
2025
Address:
Suzhou, China
Editors:
Barry Haddow, Tom Kocmi, Philipp Koehn, Christof Monz
Venue:
WMT
SIG:
Publisher:
Association for Computational Linguistics
Note:
Pages:
644–650
Language:
URL:
https://preview.aclanthology.org/ingest-emnlp/2025.wmt-1.39/
DOI:
Bibkey:
Cite (ACL):
Hayate Shiroma. 2025. SH at WMT25 General Machine Translation Task. In Proceedings of the Tenth Conference on Machine Translation, pages 644–650, Suzhou, China. Association for Computational Linguistics.
Cite (Informal):
SH at WMT25 General Machine Translation Task (Shiroma, WMT 2025)
Copy Citation:
PDF:
https://preview.aclanthology.org/ingest-emnlp/2025.wmt-1.39.pdf