English(EN) Cross-Preference Learning for Sentence-Level and Context-Aware Machine Translation

新的跨偏好学习方法提高了机器翻译质量

作者 PulseAugur 编辑部 · [1 个来源] · 2026-09-07 04:00

研究人员推出了一种名为跨偏好学习（CPL）的新型训练框架，旨在增强机器翻译模型。CPL 显式地模拟了不同句子中上下文信息变化的益处，使模型能够在上下文信息有益时自适应地利用它，而在无益时保持鲁棒性。该方法将内部条件和跨条件偏好都整合到优化目标中。使用 Qwen3-4B、Qwen3-8B 和 Llama-3-8B-Instruct 等模型进行的实验表明，在无需架构更改的情况下，翻译质量和鲁棒性得到了持续改进。 AI

影响这种新的训练框架通过更好地适应变化的上下文信息，有望带来更鲁棒、更准确的机器翻译系统。

排序理由该集群包含一篇详细介绍机器翻译新方法的学术论文。[lever_c_demoted from research: ic=1 ai=1.0]

在 arXiv cs.CL 阅读 →

AI 生成摘要 · Google Gemini · 来自 1 个来源。我们如何撰写摘要 →

本文如何被排名

Signal score: 32 / 100; Composite score across the factors below. Higher = stronger signal that this story matters right now.
Newsworthiness bucket: Tool; 该集群包含一篇详细介绍机器翻译新方法的学术论文。[lever_c_demoted from research: ic=1 ai=1.0]
Source corroboration: Single-source cluster; Only one publisher covered this so far. Single-source stories can still rank when the publisher is high-authority, but they lack cross-source corroboration.
Topics: paper, model release; Editorial topic classification. Feeds into how the story surfaces on /topic/<slug> hub pages and into the per-entity coverage mix.
AI-industry relevance: High; Clearly on-topic for AI-industry coverage.
Story freshness: Breaking (< 6h); Fresh story with cross-source coverage still developing. Ranking may shift as more sources report.

完整方法见我们的编辑标准。

报道来源 [1]

arXiv cs.CL TIER_1 English(EN) · Ying Li, Xinglin Lyu, Junhui Li, Jinlong Yang, Hengchao Shang, Min Zhang, Shimin Tao, Daimeng Wei · 2026-09-07 04:00

面向句子级和上下文感知机器翻译的跨偏好学习

arXiv:2603.25183v2 Announce Type: replace Abstract: Context-aware machine translation (MT) leverages document-level information, yet it does not consistently outperform sentence-level MT, as contextual signals are unevenly beneficial across sentences. Existing training objectives…

本文如何被排名

报道来源 [1]

面向句子级和上下文感知机器翻译的跨偏好学习

相关实体

相关话题