评测大模型评判者在多语言和对话轨迹中的可靠性缺陷。
BabelJudge: Measuring LLM-as-a-Judge Reliability Across Languages and Agent Trajectories
- 用可控扰动生成已知答案对，无需人工标注即可检测评判偏差。
- 发现模型在斯瓦希里语中可靠性降至0.550，顺序一致性仅0.480。
- 支持多语言和智能体评估，适合关注模型公平性的研究者使用。
LLM作为评判者已成为NLP流水线中可扩展评估的主流方法，但其自身存在系统性偏差，而原始准确率无法揭示这些缺陷：偏好置于位置A的回复（位置偏差）、更倾向长文本（冗余偏差），且在低资源语言中可靠性急剧下降。我们提出BabelJudge，一个开源基准与可靠性审计框架，可在不依赖人工偏好标签的情况下，测量任意评判模型的四种失效模式——位置偏差、冗余偏差、顺序不一致性和跨语言退化。核心思路是通过降级式黄金标注：从高质量参考回复出发施加可控扰动，生成一对具有已知正确标签的样本，从而消除标注成本。我们在英语、印地语、阿拉伯语和斯瓦希里语上评估Qwen2.5-7B-Instruct-4bit，发现复合偏差惩罚后的可靠性得分从印地语的0.714降至斯瓦希里语的0.550，而原始准确率（0.835 vs. 0.660）低估了这一差距。斯瓦希里语的顺序一致性跌至0.480，表明在位置互换时评判结果接近随机——此问题仅靠准确率无法察觉。我们进一步将框架扩展至智能体评估，引入九种轨迹级扰动（论点破坏、工具替换、虚构调用、步骤缺失）及三项新指标：工具准确率、幻觉检测率和轨迹长度偏差。BabelJudge以Python包形式发布，支持11种评判模型后端。代码：https://github.com/Shreyaskc/BabelJudge
原文摘要 · Abstract (English)
LLM-as-a-judge has become the dominant approach to scalable evaluation in NLP pipelines, yet judges themselves carry systematic biases that raw accuracy hides: they favor responses placed in slot A (position bias), they prefer longer responses regardless of quality (verbosity bias), and their reliability degrades sharply in lower-resource languages. We introduce BabelJudge, an open-source benchmark and reliability audit framework that measures all four failure modes -- position bias, verbosity bias, order inconsistency, and cross-lingual degradation -- on any judge model, without requiring human preference labels. The key insight is gold-labelling by degradation: starting from a high-quality reference response and applying a controlled perturbation yields a pairwise item whose gold label is known by construction, eliminating annotation cost. We evaluate Qwen2.5-7B-Instruct-4bit across English, Hindi, Arabic, and Swahili and find that our composite bias-penalised reliability score drops from 0.714 in Hindi to 0.550 in Swahili, a gap that raw accuracy (0.835 vs. 0.660) understates. Swahili order consistency collapses to 0.480, meaning judge verdicts are near-random under slot-order swaps -- a failure mode invisible to accuracy alone. We further extend the framework to agentic evaluation via nine trajectory-level perturbations (argument corruption, tool swaps, hallucinated calls, missing steps) and three new metrics: tool accuracy, hallucination detection rate, and trajectory-length bias. BabelJudge is released as a Python package supporting 11 judge backends. Code: https://github.com/Shreyaskc/BabelJudge
Thank you to arXiv for use of its open access interoperability. PaperDance 不是 arXiv 官方产品；中文卡片由大模型生成，请以原文为准。