[2602.15029] Symmetry in language statistics shapes the geometry of model representations
Abstract:The internal representations learned by language models consistently exhibit striking geometric structure: calendar months organize into a circle, historical years form a smooth one-dimensional manifold, and cities' latitudes and longitudes can be decoded using a linear probe. To explain this neural code, we first show that language statistics exhibit translation symmetry (for example, the frequency with which any two months co-occur in text depends only on the time interval between them). We prove that this symmetry governs these geometric structures in high-dimensional word embedding models, and we analytically derive the manifold geometry of word representations. These predictions empirically match large text embedding models and large language models. Moreover, the representational geometry persists at moderate embedding dimension even when the relevant statistics are perturbed (e.g., by removing all sentences in which two months co-occur). We prove that this robustness emerges naturally when the co-occurrence statistics are controlled by an underlying latent variable. Our results indicate that these representational manifolds originate in the statistical symmetries of natural language.
View PDF HTML (experimental) Abstract:The internal representations learned by language models consistently exhibit striking geometric structure: calendar months organize into a circle, historical years form a smooth one-dimensional manifold, and cities' latitudes and longitudes can be decoded using a linear probe. To explain this neural code, we first show that language statistics exhibit translation symmetry (for example, the frequency with which any two months co-occur in text depends only on the time interval between them). We prove that this symmetry governs these geometric structures…
saved by
related reading
- [2602.15029] Symmetry in language statistics shapes the geometry of model representationsarxiv.org
- Verbalizable Representations Form a Global Workspace in Language Modelstransformer-circuits.pub
- Training Language Models to Explain Their Own Computationsarxiv.org
- Interpreting Language Model Parametersarxiv.org
- The Linear Representation Hypothesis and the Geometry of Large Language Modelsarxiv.org
- The World Inside Neural Networksgoodfire.ai
- Circuit Tracing: Revealing Computational Graphs in Language Modelstransformer-circuits.pub
- Language, trees, and geometry in neural networkspair-code.github.io
- Towards Monosemanticity: Decomposing Language Models With Dictionary Learningtransformer-circuits.pub
- When Models Manipulate Manifolds: The Geometry of a Counting Tasktransformer-circuits.pub
- On the Origins of Linear Representations in Large Language Modelsarxiv.org
- Language Models, World Models, and Human Model-Buildinglingo.csail.mit.edu