How Well Do Large Reasoning Models Translate? A Comprehensive Evaluation for Multi-Domain Machine Translation

TL;DR

This study systematically evaluates Large Reasoning Models (LRMs) versus traditional LLMs for multi-domain machine translation (MDMT) across 15 domains and four translation directions, using BLEU, COMET, CometKiwi, and a domain-enhanced MQM framework. LRMs demonstrate superior semantic accuracy and document-level coherence in complex, long-form translations, while struggles remain in terminology-heavy and stylistically constrained domains, where traditional LLMs maintain lexical precision. The authors show that domain-aware prompting—especially prompting that enables domain inference—consistently improves semantic and stylistic alignment. Overall, the findings highlight the potential of structured reasoning in MDMT and suggest that integrating domain-specific constraints with LRMs could yield more robust, domain-sensitive translation systems.

Abstract

Large language models (LLMs) have demonstrated strong performance in general-purpose machine translation, but their effectiveness in complex, domain-sensitive translation tasks remains underexplored. Recent advancements in Large Reasoning Models (LRMs), raise the question of whether structured reasoning can enhance translation quality across diverse domains. In this work, we compare the performance of LRMs with traditional LLMs across 15 representative domains and four translation directions. Our evaluation considers various factors, including task difficulty, input length, and terminology density. We use a combination of automatic metrics and an enhanced MQM-based evaluation hierarchy to assess translation quality. Our findings show that LRMs consistently outperform traditional LLMs in semantically complex domains, especially in long-text and high-difficulty translation scenarios. Moreover, domain-adaptive prompting strategies further improve performance by better leveraging the reasoning capabilities of LRMs. These results highlight the potential of structured reasoning in MDMT tasks and provide valuable insights for optimizing translation systems in domain-sensitive contexts.

Figures (5)

• Figure 1: Multi-Domain En$\Leftrightarrow$De Translation Performance Comparison, showing averaged BLEU, COMET, and CometKiwi scores for both directions, with distinct colors representing different LLMs.
• Figure 2: Multi-Domain En$\Leftrightarrow$Zh Translation Performance Comparison, showing averaged BLEU, COMET, and CometKiwi scores for both directions, with distinct colors representing different LLMs.
• Figure 3: Case study of DeepSeek-R1 showing its reasoning process and final translation.
• Figure 4: Full prompt used to calculate MQM scores with DeepSeek-V3.
• Figure 5: Full prompt used for evaluating translation difficulty with DeepSeek-V3.