LLMs中的推理是否由不同的语义结构介导？一种机制性解释
Is Inference Mediated by Distinct Semantic Structures in LLMs? A Mechanistic Interpretation
AI总结：
通过SVD分解和激活引导实验，研究自然语言推理中Transformer模型是否编码语义操作，发现操作级子空间部分重叠且因果影响预测，表明模型不仅编码假设与前提的关系，还部分编码如何关联。
摘要：
正确预测标签并不一定需要表示产生该标签的操作。已知Transformer表示携带标签级信息，但它们是否编码产生这些标签的语义操作尚不清楚。我们使用受控的前提-假设对（仅通过单一语义变换区分）在自然语言推理中对此进行研究。利用逐层激活，通过SVD估计操作级子空间，并通过在四个开源解码器模型中的激活引导测试其因果相关性。变换效果以84.8%-99%的准确率可解码，并占据部分不同但重叠的子空间，超过随机子空间基线。引导实验表明这些方向因果性地影响预测，尽管可引导性因模型而异；跨操作引导进一步揭示了结构化干扰以及子空间选择性与跨操作独立性之间的分离。这些发现表明，模型不仅编码假设与前提相关，还部分编码如何相关，这意味着机制分析和控制应在语义操作层面而非仅预测标签层面进行。
英文摘要：
Predicting a label correctly does not necessarily require representing the operation that produces it. Transformer representations are known to carry label-level information, but whether they encode semantic operations producing those labels is unclear. We investigate this in Natural Language Inference using controlled premise-hypothesis pairs that differ by a single semantic transformation. Using layer-wise activations, we estimate operation-level subspaces via SVD and test their causal relevance through activation steering in four open-weight decoder models. Transformation effects are decodable with $84.8$-$99\%$ accuracy and occupy partially distinct but overlapping subspaces, exceeding random-subspace baselines. Steering experiments show that these directions causally influence predictions, though steerability varies across models; cross-operation steering further reveals structured interference and a dissociation between subspace selectivity and cross-operation independence. These findings indicate that the models encode not only that a hypothesis relates to a premise but also, in part, how it does so, implying that mechanistic analysis and control should operate at the level of semantic operations rather than predicted labels alone.
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。
