Synapse
⌘+K
Synapse
PulseExploreJournal ClubResearchersJournals
Instagram
HomeJournal ClubExplore
July 26, 2026ACM Transactions on Asian and Low-Resource Language Information ProcessingOpen Access

Semi-Supervised Marginal Likelihood Training with Curriculum-Guided Rewriting for Low-Resource Machine Translation

View Full Paper
Ask AI
Bookmark
Share

Authors

WYWenjie YuZYZhiqiang YuJZJiang Zuo

Discussion

Loading...

Member takes

Overview

Randomized trial investigates translation improvements in low-resource languages, indicating potential for effective model fine-tuning.

Key Points

  • The study aims to improve low-resource language translation by effectively fine-tuning large language models using monolingual data.
  • Developed a semi-supervised framework integrating marginal distribution estimation with curriculum-guided rewriting.
  • Conducted experiments in four low-resource translation directions and assessed performance improvements.
  • Utilized reference-free metrics to measure fluency and adequacy of translations.
  • Achieved an average increase of +8 spBLEU and +10 COMET over strong baselines in low-resource directions.
  • Showed stable improvements in three additional mid-resource directions.
  • Confirmed robust gains in translation fluency and adequacy through reference-free metrics.

Cite This Study

Yu et al. (2026) studied this question.

synapsesocial.com/papers/6a65a337d3aea3239cd76706https://doi.org/10.1145/3830466
View Full Paper
Ask AI
Bookmark
Share