BBVA/SocialStigmaQA-ES

SocialStigmaQA-ES数据集是一个针对西班牙语的社交偏见审计工具，它改编自英文版SocialStigmaQA数据集。该数据集包含大约10K个提示，这些提示结合了37种情境、93个以美国为中心的偏见标签以及三种不同的提示风格。数据集旨在评估生成式语言模型在处理社交情境时的偏见反应。The SocialStigmaQA-ES dataset is a tool for social bias auditing targeting the Spanish language, adapted from the English SocialStigmaQA dataset. It contains approximately 10K prompts that combine 37 scenarios, 93 US-centric stigma labels, and three different prompt styles. The dataset is designed to evaluate the bias responses of generative language models in handling social situations.