Are Large Reasoning Models Good Translation Evaluators? Analysis and Performance Boost
This addresses the need for efficient and accurate automatic evaluation in machine translation, though it is incremental as it builds on existing LRM capabilities.
The paper tackles the problem of using large reasoning models (LRMs) as evaluators for machine translation quality, identifying key challenges like overthinking and scoring issues, and proposes calibrating LRM thinking with synthetic human-like trajectories to reduce thinking budgets by ~35x and improve evaluation performance (e.g., +8.7 correlation points).
Recent advancements in large reasoning models (LRMs) have introduced an intermediate "thinking" process prior to generating final answers, improving their reasoning capabilities on complex downstream tasks. However, the potential of LRMs as evaluators for machine translation (MT) quality remains underexplored. We provides the first systematic analysis of LRM-as-a-judge in MT evaluation. We identify key challenges, revealing LRMs require tailored evaluation materials, tend to "overthink" simpler instances and have issues with scoring mechanisms leading to overestimation. To address these, we propose to calibrate LRM thinking by training them on synthetic, human-like thinking trajectories. Our experiments on WMT24 Metrics benchmarks demonstrate that this approach largely reduces thinking budgets by ~35x while concurrently improving evaluation performance across different LRM scales from 7B to 32B (e.g., R1-Distill-Qwen-7B achieves a +8.7 correlation point improvement). These findings highlight the potential of efficiently calibrated LRMs to advance fine-grained automatic MT evaluation.