Polar probe linearly decodes semantic structures from LLMs
Not on the map. Why?
Abstract (arXiv)
How do artificial neural networks bind concepts to form complex semantic structures? Here, we propose a simple neural code, whereby the existence and the type of relations between entities are represented by the distance and the direction between their embeddings, respectively. We test this hypothesis in a variety of Large Language Models (LLMs), each input with natural-language descriptions of minimalist tasks from five different domains: arithmetic, visual scenes, family trees, metro maps and social interactions. Results show that the true semantic structures can be linearly recovered with a Polar Probe targeting a subspace of LLMs' layer activations. Second, this code emerges mostly in middle layers and improves with LLM performance. Third, these Polar Probes successfully generalize to new entities and relation types, but degrades with the size of the semantic structure. Finally, the quality of the polar representation correlates with the LLM's ability to answer questions about the semantic structure. Together, these findings suggest that LLMs learn to build complex semantic structures by binding representations with a simple geometrical principle.
Key facts. Cited by 2 papers in the map (data through 14 Oct 2026). Builds on 27 of its 37 references that are in the map. First posted on arXiv 13 May 2026.
Builds on
27 of 37 references.
References · 27
- 2017Attention Is All You Needcited by 60,361
- 2020Language Models are Few-Shot Learnerscited by 32,419
- 2013Efficient Estimation of Word Representations in Vector Spacecited by 4,375
- 2023Pythia: A Suite for Analyzing Large Language Models Across Training and Scalingcited by 1,786
- 2016Understanding intermediate layers using linear classifier probescited by 1,353
- 2019BERT Rediscovers the Classical NLP Pipelinecited by 1,059
- 2022Toy Models of Superpositioncited by 971
- 2023Steering Language Models With Activation Engineeringcited by 922
- 2023The Geometry of Truth: Emergent Linear Structure in Large Language Model Representations of True/False Datasetscited by 712
- 2024OLMo: Accelerating the Science of Language Modelscited by 613
- 2018What you can cram into a single vector: Probing sentence embeddings for linguistic propertiescited by 457
- 2019Linguistic Knowledge and Transferability of Contextual Representationscited by 383
15 more
- 2023Emergent Linear Representations in World Models of Self-Supervised Sequence Modelscited by 319
- 2023Language Models Represent Space and Timecited by 316
- 2020Fast Differentiable Sorting and Rankingcited by 197
- 2024The Geometry of Categorical and Hierarchical Concepts in Large Language Modelscited by 159
- 2023How do Language Models Bind Entities in Context?cited by 108
- 2024ICLR: In-Context Learning of Representationscited by 50
- 2024Language Models Encode Numbers Using Digit Representations in Base 10cited by 32
- 2025Linear Representations of Political Perspective Emerge in Large Language Modelscited by 32
- 2024Representational Analysis of Binding in Language Modelscited by 28
- 2024A polar coordinate system represents syntax in large language modelscited by 25
- 2022Probing for Incremental Parse States in Autoregressive Language Modelscited by 24
- 2020Introducing Orthogonal Constraint in Structural Probescited by 21
- 2022Probing for Labeled Dependency Treescited by 8
- 2023More than Correlation: Do Large Language Models Learn Causal Representations of Space?cited by 7
- 2025Probing Syntax in Large Language Models: Successes and Remaining Challengescited by 5
Built on by
Citing papers per year: 2026: 2.
2 papers.