AlignCultura: 向文化对齐的大语言模型迈进？
AlignCultura: Towards Culturally Aligned Large Language Models?
AI总结：
本文提出AlignCultura框架，通过构建HHH-English数据集和评估不同模型的文化对齐能力，提升输出的尊重性和文化多样性。
摘要：
大语言模型中的文化对齐对于生成具有上下文意识、尊重和可信的输出至关重要。缺乏这种对齐会导致模型生成刻板印象、不敏感或误导性的回应，无法反映文化多样性相对于Helpful, Harmless, and Honest (HHH)范式。现有的基准测试是文化对齐的早期步骤；然而，目前没有基准测试能够系统评估文化对齐，符合联合国教科文组织关于文化多样性相对于HHH范式的原则。因此，为了解决这一差距，我们构建了Align-Cultura，一个两阶段的文化对齐流水线。第一阶段构建CULTURAX，通过查询构造，基于联合国教科文组织的文化分类构建HHH-English数据集，通过重新分类提示、扩展代表性不足的领域（或标签）以及使用SimHash防止数据泄漏。然后，响应生成通过两阶段拒绝采样将提示与文化扎根的响应配对。最终的数据集包含1,500个样本，涵盖30个物质和非物质文化形式的子领域。第二阶段在通用模型、文化微调模型和开放式权重LLM（Qwen3-8B和DeepSeek-R1-Distill-Qwen-7B）上评估CULTURAX。实证结果显示，文化微调模型在联合HHH方面提高了4%-6%，减少了文化失败18%，实现了10%-12%的效率提升，并将泄漏限制在0.3%。
英文摘要：
Cultural alignment in Large Language Models (LLMs) is essential for producing contextually aware, respectful, and trustworthy outputs. Without it, models risk generating stereotyped, insensitive, or misleading responses that fail to reflect cultural diversity w.r.t Helpful, Harmless, and Honest (HHH) paradigm. Existing benchmarks represent early steps toward cultural alignment; yet, no benchmarks currently enables systematic evaluation of cultural alignment in line with UNESCO's principles of cultural diversity w.r.t HHH paradigm. Therefore, to address this gap, we built Align-Cultura, two-stage pipeline for cultural alignment. Stage I constructs CULTURAX, the HHH-English dataset grounded in the UNESCO cultural taxonomy, through Query Construction, which reclassifies prompts, expands underrepresented domains (or labels), and prevents data leakage with SimHash. Then, Response Generation pairs prompts with culturally grounded responses via two-stage rejection sampling. The final dataset contains 1,500 samples spanning 30 subdomains of tangible and intangible cultural forms. Stage II benchmarks CULTURAX on general-purpose models, culturally fine-tuned models, and open-weight LLMs (Qwen3-8B and DeepSeek-R1-Distill-Qwen-7B). Empirically, culturally fine-tuned models improve joint HHH by 4%-6%, reduce cultural failures by 18%, achieve 10%-12% efficiency gains, and limit leakage to 0.3%.
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。
