纠缠于表征：大型语言模型中文化偏见的机制性探究
CulTrace: Tracing Internal Cultural Reasoning in Large Language Models
AI总结：
本研究提出Culturescope方法，通过机制性可解释性探讨LLMs中文化偏见的来源，揭示低资源文化对文化偏见的抗性及模型参数知识的限制。
摘要：
随着大型语言模型（LLMs）在多样化文化背景下的广泛应用，深入理解LLMs对不同文化的表征变得至关重要。先前的研究仅通过评估LLMs生成的文本来检验其文化意识，这种做法忽略了模型内部文化误表征的内部来源。为填补这一空白，我们提出了Culturescope，这是首个基于机制性可解释性方法，用于探测LLMs中不同文化知识的内部表征。我们还引入了一个文化扁平化分数，作为衡量从Culturescope解码知识中固有文化偏见的指标。此外，我们研究了LLMs如何内化文化偏见，这使我们能够追溯西方主导偏见和文化扁平化等文化偏见如何在LLMs中产生。我们发现低资源文化较少受到文化偏见的影响，这可能归因于模型参数知识的有限性。我们的工作为未来研究如何减轻文化偏见和增强LLMs的文化理解奠定了基础。
英文摘要：
The growing deployment of large language models (LLMs) across diverse cultural contexts necessitates a deeper understanding of models' hidden representations of different cultures. Prior work has evaluated cultural awareness in LLMs by analysing their outputs. This approach overlooks how cultures are represented within the model parameters, missing why models generate incorrect responses. To bridge this gap, we propose CulTrace, a mechanistic interpretability-based method that probes the internal representations of LLMs for cultural knowledge. With CulTrace, we inspect how cultural knowledge is processed across layers and how it is integrated during cultural QA. We find a consistent staged trajectory of cultural reasoning. Models first engage with the question's domain, then resolve the relevant culture, and finally narrow in on an answer. We also demonstrate that models' cultural reasoning is imbalanced, showing delayed relevant culture resolution and more confusion with less-represented cultures.
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。
