Cross-Preference Learning for Sentence-Level and Context-Aware Machine Translation

TL;DR

This paper introduces Cross-Preference Learning, a training framework that explicitly models when context benefits translation, leading to improved and more robust context-aware machine translation without changing model architecture.

Contribution

It proposes a preference-based training method that captures the benefits of sentence-level and context-aware MT, enabling models to adaptively leverage context.

Findings

Consistent improvements in translation quality across multiple tasks.

Enhanced robustness of models with respect to contextual information.

Effective without architectural modifications.

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 do not explicitly model this variability, limiting a model's ability to adaptively exploit context. In this paper, we propose Cross-Preference Learning (CPL), a preference-based training framework that explicitly captures the complementary benefits of sentence-level and context-aware MT. CPL achieves this by integrating both intra- and cross-condition preferences into the preference optimization objective. The introduction of intra- and cross-condition preferences provides explicit supervision on when and how contextual information improves translation quality. We validate the proposed approach on several public context-aware MT tasks using multiple models, including Qwen3-4B, Qwen3-8B, and Llama-3-8B. Experimental results demonstrate consistent improvements in translation quality and robustness across both input conditions, achieved without any architectural modifications.