BSC-LT/hhh_alignment_es

hhh_alignment_es（Helpful, Honest, & Harmless - a Pragmatic Alignment Evaluation - Spanish）是一个西班牙语的问答数据集，旨在评估语言模型在帮助性、诚实性、无害性等方面的对齐性。数据集包含61个诚实性、59个帮助性、58个无害性和43个其他类别的实例，每个实例包含一个输入提示和两个可能的答案。数据集由BSC-CNS的语言技术部门策划，并由ILENIA项目资助。数据集的创建基于Askell等人（2021）的研究，旨在通过翻译原始英文数据集来支持西班牙语的NLP研究。

hhh_alignment_es (Helpful, Honest, & Harmless - a Pragmatic Alignment Evaluation - Spanish) is a question answering dataset in Spanish designed to evaluate language models on alignment, particularly in terms of helpfulness, honesty, and harmlessness. The dataset comprises 61 honesty, 59 helpfulness, 58 harm, and 43 other categorized instances, each containing an input prompt with two possible target answers. Curated by the Language Technologies Unit at BSC-CNS and funded by the ILENIA project, the dataset is based on the research by Askell et al. (2021) and aims to support NLP research in Spanish by translating the original English dataset.