Sociocultural Statements
收藏资源简介:
该数据集名为Sociocultural Statements,由克卢日-纳波卡技术大学和亚利桑那大学联合构建,包含281条从在线辩论平台Kialo采集的自然辩论语句。数据通过多阶段标注流程进行霍夫斯泰德文化维度标定,经LLM初筛和人工验证确保质量,主要用于量化多语言大语言模型对社会文化价值观的隐含认知。其创新性在于采用自然对话场景替代传统问卷形式,为研究AI模型的文化适应性提供了更生态效度的评估工具。
This dataset, designated as Sociocultural Statements, was co-developed by the Technical University of Cluj-Napoca and the University of Arizona. It encompasses 281 natural debate statements collected from the online debate platform Kialo. The data was annotated against Hofstede's cultural dimensions via a multi-stage annotation pipeline, with initial screening conducted by large language models (LLMs) followed by manual validation to ensure data quality. It is primarily designed to quantify the implicit socio-cultural value perceptions of multilingual large language models. Its core innovation lies in replacing traditional questionnaire formats with natural conversational scenarios, providing a more ecologically valid assessment tool for researching the cultural adaptability of AI models.
- 1Understanding Cultural Alignment in Multilingual LLMs via Natural Debate Statements克卢日-纳波卡技术大学·计算机科学系; 亚利桑那大学·计算机科学系 · 2026年



