Saved in:
Bibliographic Details
Main Authors: Lee, Dongyub Jude, Ye, Zhenyi, He, Pengcheng
Format: Preprint
Published: 2025
Subjects:
Online Access:https://arxiv.org/abs/2507.22219
Tags: Add Tag
No Tags, Be the first to tag this record!
_version_ 1866915686669877248
author Lee, Dongyub Jude
Ye, Zhenyi
He, Pengcheng
author_facet Lee, Dongyub Jude
Ye, Zhenyi
He, Pengcheng
contents Preference-learning methods for machine translation (MT), such as Direct Preference Optimization (DPO), have shown strong gains but typically rely on large, carefully curated preference triplets and often struggle to generalize beyond their tuning domains. We propose Reinforcement Learning from Teacher-Model Refinement (RLfR), which replaces static triplets with on-policy, actor-conditioned refinements produced by a frozen teacher. At each step, the actor samples candidate translations, the teacher performs a minimal local edit of each draft, and the actor is reinforced to close the gap using a composite reward that combines scaled negative edit distance for lexical and structural fidelity with COMET for semantic adequacy. This formulation yields a stable, model-aware learning signal without requiring explicit preference datasets. Experiments on FLORES-200 (English to German, Spanish, Chinese, Korean, and Japanese) show that RLfR consistently outperforms strong MT-SFT, DPO, and fixed-reference RL baselines, improving semantic quality and entity preservation, and also achieves superior performance under LLM-based judge evaluations.
format Preprint
id arxiv_https___arxiv_org_abs_2507_22219
institution arXiv
publishDate 2025
record_format arxiv
spellingShingle RL from Teacher-Model Refinement: Gradual Imitation Learning for Machine Translation
Lee, Dongyub Jude
Ye, Zhenyi
He, Pengcheng
Computation and Language
Artificial Intelligence
Preference-learning methods for machine translation (MT), such as Direct Preference Optimization (DPO), have shown strong gains but typically rely on large, carefully curated preference triplets and often struggle to generalize beyond their tuning domains. We propose Reinforcement Learning from Teacher-Model Refinement (RLfR), which replaces static triplets with on-policy, actor-conditioned refinements produced by a frozen teacher. At each step, the actor samples candidate translations, the teacher performs a minimal local edit of each draft, and the actor is reinforced to close the gap using a composite reward that combines scaled negative edit distance for lexical and structural fidelity with COMET for semantic adequacy. This formulation yields a stable, model-aware learning signal without requiring explicit preference datasets. Experiments on FLORES-200 (English to German, Spanish, Chinese, Korean, and Japanese) show that RLfR consistently outperforms strong MT-SFT, DPO, and fixed-reference RL baselines, improving semantic quality and entity preservation, and also achieves superior performance under LLM-based judge evaluations.
title RL from Teacher-Model Refinement: Gradual Imitation Learning for Machine Translation
topic Computation and Language
Artificial Intelligence
url https://arxiv.org/abs/2507.22219