ACL Findings2024
Likelihood-based Mitigation of Evaluation Bias in Large Language Models
Masanari Ohi, Masahiro Kaneko, Ryuto Koike, Mengsay Loem, Naoaki Okazaki
Abstract
Large Language Models (LLMs) are widely used to evaluate natural language generation tasks as automated metrics. However, the likelihood, a measure of LLM's plausibility for a sentence, can vary due to superficial differences in sentences, such as word order and sentence structure. It is therefore possible that there might be a likelihood bias if LLMs are used for evaluation: they might overrate sentences with higher likelihoods while underrating those with lower likelihoods. In this paper, we investigate the presence and impact of likelihood bias in LLM-based evaluators. We also propose a method to mitigate the likelihood bias.
Our method utilizes highly biased instances as few-shot examples for in-context learning. Our experiments in evaluating the data-to-text and grammatical error correction tasks reveal that several LLMs we test display a likelihood bias. Furthermore, our proposed method successfully mitigates this bias, also improving evaluation performance (in terms of correlation of models with human scores) significantly.
Ask Lune
Know this paper in minutes, not an afternoon.
Lune can read every page of it and the top-tier papers that cite it, and gives you its claims, its evidence and its limits, each one cited to the paper it comes from. No skimming, no AI slop.
7 top-tier papers have cited it since. Know what they found before your reviewers do.
Lune can draw on
the full paper · the top-tier papers that cite it · related work
Free to start. No credit card required.
Lineage
- From Generation to Judgment: Opportunities and Challenges of LLM-as-a-judge cites G-Eval: NLG Evaluation using Gpt-4 with Better Human Alignment
- From Generation to Judgment: Opportunities and Challenges of LLM-as-a-judge cites Chain-of-Thought Prompting Elicits Reasoning in Large Language Models
- Assessing the Belief Consistency of Large Language Models on the Logical Conversation Process cites Chain-of-Thought Prompting Elicits Reasoning in Large Language Models
- G-Eval: NLG Evaluation using Gpt-4 with Better Human Alignment cites Chain-of-Thought Prompting Elicits Reasoning in Large Language Models
Cited by top-tier papers7
- From Generation to Judgment: Opportunities and Challenges of LLM-as-a-judgeDawei Li, Bohan Jiang, Liangjie Huang, Alimohammad Beigi et al.EMNLP 2025 · 37 citations
- Rolling the DICE on Idiomaticity: How LLMs Fail to Grasp ContextMaggie Mi, Aline Villavicencio, Nafise Sadat MoosaviACL 2025 · 9 citations
- JailNewsBench: Multi-Lingual and Regional Benchmark for Fake News Generation under Jailbreak AttacksMasahiro Kaneko, Ayana Niwa, Timothy BaldwinICLR 2026 · 6 citations
- A Multilingual Social Bias Benchmark Incorporating Thinking ProcessesMasahiro Kaneko, Danushka Bollegala, Timothy BaldwinACL 2026
- Assessing the Belief Consistency of Large Language Models on the Logical Conversation ProcessTomoki Tsujimura, Matiss Rikters, Masaki Asada, Shusaku Egami et al.ACL 2026
Builds on5
- Chain-of-Thought Prompting Elicits Reasoning in Large Language ModelsJason Wei, Xuezhi Wang, Dale Schuurmans, Maarten Bosma et al.NeurIPS 2022 · 22,562 citations
- G-Eval: NLG Evaluation using Gpt-4 with Better Human AlignmentYang Liu, Dan Iter, Yichong Xu, Shuohang Wang et al.EMNLP 2023 · 549 citations
- Evaluating Open-Domain Question Answering in the Era of Large Language ModelsEhsan Kamalloo, Nouha Dziri, Charles L. A. Clarke, Davood RafieiACL 2023 · 96 citations
- Language Models as an Alternative Evaluator of Word Order Hypotheses: A Case Study in JapaneseTatsuki Kuribayashi, Takumi Ito, Jun Suzuki, Kentaro InuiACL 2020 · 2 citations
- OpenMEVA: A Benchmark for Evaluating Open-ended Story Generation MetricsJian Guan, Zhexin Zhang, Zhuoer Feng, Zitao Liu et al.ACL 2021
Related papers
- Rating Roulette: Self-Inconsistency in LLM-As-A-Judge FrameworksRajarshi Haldar, Julia HockenmaierEMNLP Findings 2025 · 11 citations
- Social Bias Evaluation for Large Language Models Requires Prompt VariationsRem Hida, Masahiro Kaneko, Naoaki OkazakiEMNLP Findings 2025 · 7 citations
- CalibraEval: Calibrating Prediction Distribution to Mitigate Selection Bias in LLMs-as-JudgesHaitao Li, Junjie Chen, Qingyao Ai, Zhumin Chu et al.ACL 2025
- Co-Eval: Augmenting LLM-based Evaluation with Machine MetricsLing-I Wu, Weijie Wu, Minyu Chen, Jianxin Xue et al.EMNLP 2025
- Quantifying Biases in LLM-as-a-Judge EvaluationsMagda Dubois, Harry Coppock, Mario Giulianelli, Ole Jorgensen et al.ICML 2026
