Saved in:
| Main Authors: | , , , , , , |
|---|---|
| Format: | Preprint |
| Published: |
2026
|
| Subjects: | |
| Online Access: | https://arxiv.org/abs/2601.10187 |
| Tags: |
Add Tag
No Tags, Be the first to tag this record!
|
| _version_ | 1866915758430224384 |
|---|---|
| author | Cui, Ziang Yu, Mengran Li, Tianjiao Shi, Chenyu Shi, Yingxuan Zhang, Lusheng Lin, Hongwei |
| author_facet | Cui, Ziang Yu, Mengran Li, Tianjiao Shi, Chenyu Shi, Yingxuan Zhang, Lusheng Lin, Hongwei |
| contents | Large Language Models (LLMs) have achieved remarkable strides in multilingual translation but are hindered by a systemic cross-lingual verbosity bias, rendering them unsuitable for strict time-constrained tasks like subtitling and dubbing. Current prompt-engineering approaches struggle to resolve this conflict between semantic fidelity and rigid temporal feasibility. To bridge this gap, we first introduce Sand-Glass, a benchmark specifically designed to evaluate translation under syllable-level duration constraints. Furthermore, we propose HOMURA, a reinforcement learning framework that explicitly optimizes the trade-off between semantic preservation and temporal compliance. By employing a KL-regularized objective with a novel dynamic syllable-ratio reward, HOMURA effectively "tames" the output length. Experimental results demonstrate that our method significantly outperforms strong LLM baselines, achieving precise length control that respects linguistic density hierarchies without compromising semantic adequacy. |
| format | Preprint |
| id |
arxiv_https___arxiv_org_abs_2601_10187 |
| institution | arXiv |
| publishDate | 2026 |
| record_format | arxiv |
| spellingShingle | HOMURA: Taming the Sand-Glass for Time-Constrained LLM Translation via Reinforcement Learning Cui, Ziang Yu, Mengran Li, Tianjiao Shi, Chenyu Shi, Yingxuan Zhang, Lusheng Lin, Hongwei Computation and Language Artificial Intelligence Large Language Models (LLMs) have achieved remarkable strides in multilingual translation but are hindered by a systemic cross-lingual verbosity bias, rendering them unsuitable for strict time-constrained tasks like subtitling and dubbing. Current prompt-engineering approaches struggle to resolve this conflict between semantic fidelity and rigid temporal feasibility. To bridge this gap, we first introduce Sand-Glass, a benchmark specifically designed to evaluate translation under syllable-level duration constraints. Furthermore, we propose HOMURA, a reinforcement learning framework that explicitly optimizes the trade-off between semantic preservation and temporal compliance. By employing a KL-regularized objective with a novel dynamic syllable-ratio reward, HOMURA effectively "tames" the output length. Experimental results demonstrate that our method significantly outperforms strong LLM baselines, achieving precise length control that respects linguistic density hierarchies without compromising semantic adequacy. |
| title | HOMURA: Taming the Sand-Glass for Time-Constrained LLM Translation via Reinforcement Learning |
| topic | Computation and Language Artificial Intelligence |
| url | https://arxiv.org/abs/2601.10187 |