Rubric-as-Experts: Case-Specific MQM Rubrics for Translation Quality Evaluation

The article presents a novel framework for translation quality evaluation that utilizes case-specific dynamic rubrics within the Multidimensional Quality Metrics (MQM) taxonomy. This approach allows for adaptive allocation of evaluation spaces tailored to the complexity and granularity of individual translation instances, leading to improved Matthews Correlation Coefficient (MCC) scores and more accurate span-level error localization on WMT benchmarks. This advancement is significant for practitioners as it enhances the precision of fine-grained translation quality assessments, addressing limitations of static rubric configurations in large language model applications.