BenchMarker: An Education-Inspired Toolkit for Highlighting Flaws in Multiple-Choice Benchmarks
Nishant Balepur, Bhavya Rajasekaran, Hyunjin Jane Oh, Michael Xie, Atrey Desai, Vipul Gupta, Steven James Moore, Eunsol Choi, Rachel Rudinger, Jordan Lee Boyd-Graber
摘要
Multiple-choice question answering (MCQA) is standard in NLP, but benchmarks lack rigorous quality control. We present BenchMarker, an education-inspired toolkit using LLM judges to flag three common MCQ flaws: 1) contamination-items appearing exactly online; 2) shortcuts-cues in the choices that enable guessing; and 3) writing errors-structural/grammatical issues based on a 19-rule education rubric. We validate BenchMarker with human annotations, then run the tool to audit 12 benchmarks, revealing: 1) flaws persist in MCQA benchmarks, especially automatically-made and crowdsourced data-we detect 47% of TruthfulQA appears online and 100% of HellaSwag violates multiple writing rules; 2) contaminated MCQs tend to inflate accuracy, while writing errors tend to lower it and change rankings beyond random; and 3) prior benchmark repairs address their targeted issues (i.e., lowering accuracy with LLMwritten distractors), but inadvertently add new flaws (i.e. implausible distractors, many correct answers). Overall, flaws in MCQs degrade NLP evaluation, but education research offers a path forward. We release BenchMarker to bridge the fields and improve MCQA benchmark design. 1 Multiple-Choice Question j MCQ j … MCQ j … MCQ j … The appropriate place… www.quizlet.com Which can you drink? Inferred Question Rule 7: Is grammar consistent? 19-Rule Rubric from Education 1 (no question match) 0 (matches question) 1 (violates rule) 0 (no rule violation) Guess w/out question 1 (exists online) 0 (not online) + Contamination ( §3.1) Shortcuts ( §3.2) Writing Errors ( §3.3) Web search Report Card Contamination: 0.21 > Item j was found online Shortcuts: 0.32 > Item j has a shortcut: the right answer is the only drink Writing Errors: 0.45 > Item j R7: "This item" has a grammar mismatch -"plates" MCQA Benchmark BenchMarker: Highlighting Flaws Question: The appropriate place to put this item is a recycling bin Choices: (A) used motor oil (B) used soda can (C) used styrofoam plates (D) left over medicine Answer: (B)
问问这篇 Paper
智能体会读完全文。
Lune 把这篇 Paper 索引到了最后一个公式,引用它的顶会 Paper 也一样。你提问,回答直接引用原文。
它引用的顶会 Paper20
- Chain-of-Thought Prompting Elicits Reasoning in Large Language ModelsJason Wei, Xuezhi Wang, Dale Schuurmans, Maarten Bosma 等NeurIPS 2022 · 被引用 22,562 次
- Measuring Massive Multitask Language UnderstandingDan Hendrycks, Collin Burns, Steven Basart, Andy Zou 等ICLR 2021 · 被引用 7,905 次
- TruthfulQA: Measuring How Models Mimic Human FalsehoodsStephanie Lin, Jacob Hilton, Owain EvansACL 2022 · 被引用 3,228 次
- QASC: A Dataset for Question Answering via Sentence CompositionTushar Khot, Peter Clark, Michal Guerquin, Peter Jansen 等AAAI 2020 · 被引用 387 次
- CRUXEval: A Benchmark for Code Reasoning, Understanding and ExecutionAlex Gu, Baptiste Rozière, Hugh James Leather, Armando Solar-Lezama 等ICML 2024 · 被引用 270 次
相关 Paper
- Which of These Best Describes Multiple Choice Evaluation with LLMs? A) Forced B) Flawed C) Fixable D) All of the AboveNishant Balepur, Rachel Rudinger, Jordan Lee Boyd-GraberACL 2025
- Does Question Really Matter? The Attribution of Answer Bias in LLM EvaluationBoxi Cao, Ruotong Pan, Hongyu Lin, Xianpei Han 等AAAI 2026
- MMLU-CF: A Contamination-free Multi-task Language Understanding BenchmarkQihao Zhao, Yangyu Huang, Tengchao Lv, Lei Cui 等ACL 2025 · 被引用 33 次
- Test of Time: Rethinking Temporal Signal of Benchmark ContaminationTerry Jingchen Zhang, Gopal Dev, Ning Wang, Max Obreiter 等ACL 2026 · 被引用 3 次
- Beyond Problem Solving: UOJ-Bench for Evaluating Code Generation, Hacking, and Repair in Competitive ProgrammingTingqiang Xu, Hangrui Zhou, Tianle Cai, Alex Gu 等ICML 2026 · 被引用 1 次
