BBQ: A Hand-Built Bias Benchmark for Question Answering
Alicia Parrish,1 Angelica Chen,2 Nikita Nangia,2 Vishakh Padmakumar,2 Jason Phang,2 Jana Thompson,2 Phu Mon Htut,2 Samuel R. Bowman1,2,3 \AND1New York University Dept. of Linguistics、2New York University Center for Data Science、3New York University Dept. of Computer Science \ANDCorrespondence: {alicia.v.parrish,
Abstract
It is well documented that NLP models learn social biases, but little work has been done on how these biases manifest in model outputs for applied tasks like question answering (QA). We introduce the Bias Benchmark for QA (BBQ), a dataset of question sets constructed by the authors that highlight attested social biases against people belonging to protected classes along nine social dimensions relevant for U.S. English-speaking contexts. Our task evaluates model responses at two levels: (i) given an under-informative context, we test how strongly responses reflect social biases, and (ii) given an adequately informative context, we test whether the model’s biases override a correct answer choice. We find that models often rely on stereotypes when the context is under-informative, meaning the model’s outputs consistently reproduce harmful biases in this setting. Though models are more accurate when the context provides an informative answer, they still rely on stereotypes and average up to 3.4 percentage points higher accuracy when the correct answer aligns with a social bias than when it conflicts, with this difference widening to over 5 points on examples targeting gender for most m
中文速览
自然语言处理（NLP）模型会从训练数据中学到社会偏见，但这些偏见究竟如何在问答（QA）任务的实际输出中体现，此前研究甚少。为此，作者构建了一个名为 BBQ（Bias Benchmark for QA）的专项评测数据集，覆盖与美国英语语境相关的九个社会维度（如年龄、性别、种族、宗教等），共超过 5.8 万个手工编写的例题，每道题都针对有据可查的有害刻板印象。评测分两层进行：在信息不足的情境下，测模型会不会用刻板印象填补空白；在信息充分的情境下，测模型的偏见会不会压过文本中的正确答案。实验发现，UnifiedQA、RoBERTa 和 DeBERTaV3 等主流模型在信息不足时普遍倾向于给出符合刻板印象的错误答案，而即便在信息充分时，当正确答案与刻板印象一致时模型的准确率平均也要高出约 3.4 个百分点，在性别相关题目上差距甚至超过 5 个百分点。这一工作为系统性评估和推动 QA 模型去偏见提供了可靠的基准工具。
原文 arXiv:2110.08193;中英对照 + 大白话阅读 https://aha.fim.ai/paper/2110.08193v2