2605.29502v1 May 28, 2026 cs.CL

Source-Grounded Semantic Reinforcement Learning for Low-Resource Target-Language Generation

Wentao Zhang
Wentao Zhang
Citations: 20
h-index: 3
Xiaolu Zhang
Xiaolu Zhang
Citations: 1,075
h-index: 5
Ziyin Zhang
Ziyin Zhang
Citations: 404
h-index: 9
Dehan Li
Dehan Li
Citations: 1
h-index: 1
Zhankai Xu
Zhankai Xu
Citations: 42
h-index: 3
Zeli Su
Zeli Su
Citations: 7
h-index: 1
Zewei Pan
Zewei Pan
Citations: 15
h-index: 1
Zhouwu Liu
Zhouwu Liu
Citations: 0
h-index: 0
Di Huang
Di Huang
Citations: 54
h-index: 3
Longfei Zheng
Longfei Zheng
Ant Financial Services Group
Citations: 466
h-index: 11
Jun Zhou
Jun Zhou
Citations: 1,106
h-index: 5

Low-resource target-language generation is often limited by scarce parallel data, while high-resource source-language monolingual data is abundant but difficult to use with standard supervised fine-tuning. We propose Source-Grounded Semantic Reinforcement Learning (SG-SRL), a resource-utilization framework that converts source-language monolingual data into cross-lingual semantic supervision for target-language generation. SG-SRL performs reference-free reinforcement learning (RL) on source-language data using a cross-lingual semantic reward model, instantiated by a cross-lingual reranker that scores the semantic relevance between the source input and the target-language generation. While this induces severe verbosity-based reward hacking, a lightweight recovery stage using a small parallel corpus restores fluency, conciseness, and task format while preserving the semantic gains. Experiments on Chinese-to-Thai generation show that SG-SRL improves semantic grounding and factual coverage over cold-start SFT. Additional analyses on long-form transfer and Tibetan embedding-based rewards clarify the generalization behavior of SG-SRL and show that an encoder-based semantic reward can substitute for an LLM-based reranker in a realistic low-resource language setting.

0 Citations
0 Influential
5.5 Altmetric
27.5 Score
Original PDF

No Analysis Report Yet

This paper hasn't been analyzed by Gemini yet.

Log in to request an AI analysis.

댓글

댓글을 작성하려면 로그인하세요.

아직 댓글이 없습니다. 첫 번째 댓글을 남겨보세요!