大型语言模型在用于文本标注时会再现种族刻板印象
Large Language Models Reproduce Racial Stereotypes When Used for Text Annotation
AI总结：
研究发现大型语言模型在文本标注中会因身份线索产生系统性偏见，再现种族刻板印象，尤其在名字相关任务中表现显著。
摘要：
大型语言模型（LLMs）在从学术研究到内容审核和招聘等任务中被越来越多地用于自动化文本标注。在19个LLM和两个实验中，总计超过400万条标注判断显示，文本中微妙的身份线索会系统性地偏转标注结果，以反映种族刻板印象。在基于名字的实验中，包含与黑人相关名字的文本被18/19个模型评为更具有攻击性，18/19个模型评为更八卦。亚洲名字产生竹 ceiling 效应：17/19个模型认为个体更聪明，而18/19个模型认为他们更不自信和不社交。阿拉伯名字引发认知提升与人际贬低，四个少数族裔群体都被持续评为自我约束力较低。在匹配方言实验中，相同句子在非标准美式英语（AAVE）而非标准美式英语中被显著评为不专业（平均差距-0.774），不表明受过教育的说话者（-0.688），更具毒性（18/19）和愤怒（19/19）。值得注意的是，名称基于的招聘性中，微调似乎过度纠正，系统性地偏袒少数族裔命名的申请人。这些发现表明，使用LLMs作为自动化标注器可能会将社会模式的偏见直接嵌入到日益成为研究、治理和决策基础的数据集和测量中。
英文摘要：
Large language models (LLMs) are increasingly used for automated text annotation in tasks ranging from academic research to content moderation and hiring. Across 19 LLMs and two experiments totaling more than 4 million annotation judgments, we show that subtle identity cues embedded in text systematically bias annotation outcomes in ways that mirror racial stereotypes. In a names-based experiment spanning 39 annotation tasks, texts containing names associated with Black individuals are rated as more aggressive by 18 of 19 models and more gossipy by 18 of 19. Asian names produce a bamboo-ceiling profile: 17 of 19 models rate individuals as more intelligent, while 18 of 19 rate them as less confident and less sociable. Arab names elicit cognitive elevation alongside interpersonal devaluation, and all four minority groups are consistently rated as less self-disciplined. In a matched dialect experiment, the same sentence is judged significantly less professional (all 19 models, mean gap $-0.774$), less indicative of an educated speaker ($-0.688$), more toxic (18/19), and more angry (19/19) when written in African American Vernacular English rather than Standard American English. A notable exception occurs for name-based hireability, where fine-tuning appears to overcorrect, systematically favoring minority-named applicants. These findings suggest that using LLMs as automated annotators can embed socially patterned biases directly into the datasets and measurements that increasingly underpin research, governance, and decision-making.
机构、总结、中文摘要均由AI分析而来，可能存在错误，请以论文正文为准。
