Beneath the Surface of Chains-of-Thought: A Mechanistic Interpretation of Reasoning Operations in LLMs

• URL: http://arxiv.org/abs/2609.04753v1
• Date: Fri, 04 Sep 2026 05:37:12 GMT
• Title: Beneath the Surface of Chains-of-Thought: A Mechanistic Interpretation of Reasoning Operations in LLMs
• Abstract summary: We investigate whether distinct reasoning operations exhibit corresponding geometric structure in hidden representations.<n>We find that operations are separable in held-out representations, with separability peaking in middle layers.<n>Our work demonstrates that language models maintain representational correspondence between linguistic reasoning expressions and their internal geometric structures.
• Score: 46.97620918293402
• License: http://creativecommons.org/licenses/by/4.0/
• Abstract: Reasoning in large language models unfolds through diverse functional operations, such as problem formulation, goal decomposition, and deduction. Although these operations are explicitly distinguished in text, little is known about how they are geometrically organized in representation spaces. To this end, we investigate whether distinct reasoning operations exhibit corresponding geometric structure in hidden representations. We find that operations are separable in held-out representations, with separability peaking in middle layers, and verify that this structure is not explained by lexical or positional confounds. Across layers, token-wise operation-alignment becomes more distributed over spans, while identical surface tokens are represented differently depending on the operation of its surrounding chunk. Attention-masking interventions further show that operation-aligned representations at chunk onset depend on preceding reasoning context. Consequently, our work demonstrates that language models maintain representational correspondence between linguistic reasoning expressions and their internal geometric structures. Code and project materials are available at https://github.com/naver-ai/beneath-cot.

Related papers

• Emergent Structured Representations Support Flexible In-Context Inference in Large Language Models [77.98801218316505]
  Large language models (LLMs) exhibit emergent behaviors suggestive of human-like reasoning.<n>We investigate the internal processing of LLMs during in-context concept inference.
  arXiv  Detail & Related papers  (2026-02-08T03:14:39Z)
• The Representational Geometry of Number [1.5994376682356057]
  We show that number representations preserve a stable relational structure across tasks.<n>We find that task-specific representations are embedded in distinct subspaces, with low-level features like magnitude encoded along separable linear directions.<n>It suggests that understanding arises when task-specific transformations are applied to a shared underlying relational structure of conceptual representations.
  arXiv  Detail & Related papers  (2026-02-06T16:35:22Z)
• Steering Embedding Models with Geometric Rotation: Mapping Semantic Relationships Across Languages and Models [2.3204178451683264]
  We introduce Rotor-Invariant Shift Estimation (RISE), a geometric approach that represents semantic transformations as consistent rotational operations in embedding space.<n>RISE operations have the ability to operate across both languages and models with high transfer of performance.<n>This work provides the first systematic demonstration that discourse-level semantic transformations correspond to consistent geometric operations in multilingual embedding spaces.
  arXiv  Detail & Related papers  (2025-10-10T18:51:32Z)
• The Origins of Representation Manifolds in Large Language Models [52.68554895844062]
  We show that cosine similarity in representation space may encode the intrinsic geometry of a feature through shortest, on-manifold paths.<n>The critical assumptions and predictions of the theory are validated on text embeddings and token activations of large language models.
  arXiv  Detail & Related papers  (2025-05-23T13:31:22Z)
• How Do Transformers Learn Topic Structure: Towards a Mechanistic Understanding [56.222097640468306]
  We provide mechanistic understanding of how transformers learn "semantic structure" We show, through a combination of mathematical analysis and experiments on Wikipedia data, that the embedding layer and the self-attention layer encode the topical structure.
  arXiv  Detail & Related papers  (2023-03-07T21:42:17Z)
• Linear Spaces of Meanings: Compositional Structures in Vision-Language Models [110.00434385712786]
  We investigate compositional structures in data embeddings from pre-trained vision-language models (VLMs) We first present a framework for understanding compositional structures from a geometric perspective. We then explain what these structures entail probabilistically in the case of VLM embeddings, providing intuitions for why they arise in practice.
  arXiv  Detail & Related papers  (2023-02-28T08:11:56Z)
• The Low-Dimensional Linear Geometry of Contextualized Word Representations [27.50785941238007]
  We study the linear geometry of contextualized word representations in ELMO and BERT. We show that a variety of linguistic features are encoded in low-dimensional subspaces.
  arXiv  Detail & Related papers  (2021-05-15T00:58:08Z)
• Hierarchical Pyramid Representations for Semantic Segmentation [0.0]
  We learn the structures of objects and the hierarchy among objects because context is based on these intrinsic properties. In this study, we design novel hierarchical, contextual, and multiscale pyramidal representations to capture the properties from an input image. Our proposed method achieves state-of-the-art performance in PASCAL Context.
  arXiv  Detail & Related papers  (2021-04-05T06:39:12Z)
• Unsupervised Distillation of Syntactic Information from Contextualized Word Representations [62.230491683411536]
  We tackle the task of unsupervised disentanglement between semantics and structure in neural language representations. To this end, we automatically generate groups of sentences which are structurally similar but semantically different. We demonstrate that our transformation clusters vectors in space by structural properties, rather than by lexical semantics.
  arXiv  Detail & Related papers  (2020-10-11T15:13:18Z)

This list is automatically generated from the titles and abstracts of the papers in this site.

This site does not guarantee the quality of this site (including all information) and is not responsible for any consequences.