Unlocking Fine-Grained Translation Quality Estimation in LRMs through Synergistically Evolving Implicit and Explicit Reasoning
This work provides a novel training framework for improving fine-grained translation quality estimation in Large Reasoning Models, which is an incremental improvement for the machine translation community.
This paper addresses the challenge of fine-grained translation quality estimation (QE) in Large Reasoning Models (LRMs) by proposing RIEQE, a two-stage training framework. RIEQE synergistically evolves implicit (layer-wise) and explicit (token-wise) reasoning, achieving state-of-the-art explicit reasoning performance on WMT test sets with Qwen3-4B-Thinking-2507, while its implicit reasoning capability is comparable to the best encoder-based models.
Large Reasoning Models (LRMs) still struggle with fine-grained translation quality estimation (QE), even with long reasoning chains. We argue that LRMs already possess strong multilingual capabilities, while the core challenge stems from the intrinsic difficulty of learning the fine-grained QE task. In this paper, we propose RIEQE (Reasoning both Implicitly and Explicitly for QE), a simple two-stage training framework that enables the co-evolution of implicit (layer-wise) and explicit (token-wise) reasoning capabilities. To make implicit reasoning feasible, we first decompose the complex QE task into straightforward subtasks. Based on this, our two-stage approach applies: (1) NonThinking-SFT, Supervised Fine-Tuning (SFT) without reasoning chains to directly boost the model's implicit reasoning tendency and capability; and (2) Thinking-RLVR, standard Reinforcement Learning with Verifiable Reward (RLVR) to subsequently strengthen explicit reasoning. Results demonstrate that implicit and explicit reasoning synergistically co-evolve under our framework. On the WMT test sets, RIEQE based on Qwen3-4B-Thinking-2507 surpasses all baselines in explicit reasoning performance, while its implicit reasoning capability is also comparable to the best current encoder-based models. We further provide evidence for the synergistic collaboration between implicit and explicit reasoning, showing how they mutually benefit each other.