Translation with Thought: Difficulty-Adaptive Reasoning via Reinforcement Learning for Multi-Domain Machine Translation

2026-07-31

AI Summary

This work addresses the performance instability in multilingual machine translation caused by varying linguistic complexities across languages. The authors propose the Think-with-Task (TwT) framework, which introduces the principle of resource rationality from cognitive science into machine translation for the first time, enabling a difficulty-adaptive inference mechanism. This approach first employs a difficulty-aware chain-of-thought strategy for supervised fine-tuning and subsequently refines translation quality and inference efficiency through hybrid-reward reinforcement learning. Evaluated across 15 benchmarks covering 62 languages, TwT significantly outperforms larger state-of-the-art models while reducing inference token consumption by 32%–60% without compromising—indeed, while improving—translation quality.

Abstract

Multi-domain machine translation (MDMT) poses a unique challenge due to varying levels of linguistic complexity across domains. Inspired by human translators' ability to adapt reasoning effort based on difficulty, we propose TwT (Translation with Thought), a resource-rational framework that learns to modulate inference between intuitive and deliberate reasoning. TwT is trained in two stages: (1) supervised fine-tuning on difficulty-aware long chain-of-thought traces distilled from DeepSeek-R1 and rewritten by GPT-4o to reflect human-like reasoning economy, and (2) reinforcement learning with a hybrid reward to optimize translation quality and reasoning efficiency. Evaluated on 15 benchmarks spanning in-domain and out-of-domain settings, as well as 3 seen and 59 unseen languages, with ablations across three backbone models, TwT-7B and TwT-14B outperform much larger SOTA reasoning models in translation quality, while reducing token usage by 32--60\%. These results confirm that aligning translation behavior with cognitive principles enables robust generalization, high translation quality, and efficient reasoning in MDMT.

Problem

• multi-domain machine translation
• reasoning adaptation
• translation difficulty
• resource-rational reasoning
• linguistic complexity

Innovation

• difficulty-adaptive reasoning
• resource-rational translation
• reinforcement learning
• chain-of-thought distillation
• multi-domain machine translation