NORMDIAL

NORMDIAL是一个高质量的双语合成对话数据集，用于模拟中英文文化中的社会规范遵守与违反。该数据集由哥伦比亚大学计算机科学系创建，包含4231条对话，总计29550个对话轮次。数据集通过人类在环的流程，利用大型语言模型生成，每个对话轮次都标注了社会规范遵守或违反的情况。NORMDIAL旨在解决跨文化交流工具开发中的数据稀缺问题，特别是在识别多文化背景下的社会规范遵守与违反方面。数据集的应用领域包括提升计算系统在跨文化语境中的推理能力，促进更流畅的跨文化对话交流。

NORMDIAL is a high-quality bilingual synthetic dialogue dataset designed to simulate social norm compliance and violation in Chinese and English cultures. Developed by the Department of Computer Science, Columbia University, this dataset contains 4231 dialogues, totaling 29550 dialogue turns. It is generated via a human-in-the-loop pipeline leveraging large language models (LLMs), with each dialogue turn annotated for social norm compliance or violation. NORMDIAL aims to address the data scarcity issue in the development of cross-cultural communication tools, particularly for identifying social norm compliance and violation in multicultural contexts. The application scenarios of this dataset include enhancing the reasoning capabilities of computing systems in cross-cultural contexts and facilitating smoother cross-cultural conversational exchanges.