Polar probe linearly decodes semantic structures from LLMs

• URL: http://arxiv.org/abs/2605.14125v2
• Date: Sat, 16 May 2026 12:30:02 GMT
• Title: Polar probe linearly decodes semantic structures from LLMs
• Abstract summary: 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.<n>We test this hypothesis in a variety of Large Language Models (LLMs)<n>Results show that the true semantic structures can be linearly recovered with a Polar Probe targeting a subspace of LLMs' layer activations.
• Score: 8.391571644913968
• License: http://creativecommons.org/licenses/by/4.0/
• Abstract: 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.

Related papers

• Speaking the Language of Science: Toward a General-Purpose Generative Foundation Model for the Natural Sciences [89.71422932447423]
  LOGOS is a scientific generative language model that unifies heterogeneous tasks across the natural sciences.<n>By representing spatial contact and constraint patterns as discrete tokens, the model captures complex structural interactions in a purely sequential manner.<n>Across diverse tasks, LOGOS consistently matches or outperforms domain-specific baselines.
  arXiv  Detail & Related papers  (2026-06-15T16:14:53Z)
• Diagnosing CFG Interpretation in LLMs [17.612748633265202]
  Given a novel context-free grammar, can LLMs generate syntactically valid, behaviorally functional, and semantically faithful outputs?<n>We introduce RoboGrid, a framework that disentangles syntax, behavior, and semantics through controlled stress-tests of recursion depth, expression complexity, and surface styles.
  arXiv  Detail & Related papers  (2026-04-22T17:43:05Z)
• Improving LLM Reasoning with Homophily-aware Structural and Semantic Text-Attributed Graph Compression [55.51959317490934]
  Large language models (LLMs) have demonstrated promising capabilities in Text-Attributed Graph (TAG) understanding.<n>We argue that graphs inherently contain rich structural and semantic information, and that their effective exploitation can unlock potential gains in LLMs reasoning performance.<n>We propose Homophily-aware Structural and Semantic Compression for LLMs (HS2C), a framework centered on exploiting graph homophily.
  arXiv  Detail & Related papers  (2026-01-13T03:35:18Z)
• JE-IRT: A Geometric Lens on LLM Abilities through Joint Embedding Item Response Theory [4.680675835708409]
  We present JE-IRT, a geometric item-response framework that embeds both LLMs and questions in a shared space.<n>For question embeddings, the direction encodes semantics and the norm encodes difficulty, while correctness on each question is determined by the geometric interaction between the model and question embeddings.
  arXiv  Detail & Related papers  (2025-09-26T20:11:47Z)
• Topos Theory for Generative AI and LLMs [3.0316063849624477]
  We propose the design of novel categorical generative AI architectures (GAIAs) using topos theory.<n>We show that the category of large language models (LLMs) forms a topos, a set-like" category.<n>We use a functorial characterization of backpropagation to define a potential implementation of an LLM topos architecture.
  arXiv  Detail & Related papers  (2025-08-05T20:00:06Z)
• Meaning-infused grammar: Gradient Acceptability Shapes the Geometric Representations of Constructions in LLMs [0.8594140167290097]
  This study investigates whether the internal representations in Large Language Models (LLMs) reflect the proposed function-infused gradience.<n>We analyze representations of the English Double Object (DO) and Prepositional Object (PO) constructions in Pythia-$1.4$B.<n> Geometric analyses show that the separability between the two constructions' representations is systematically modulated by gradient preference strength.
  arXiv  Detail & Related papers  (2025-07-29T23:39:21Z)
• SAFT: Structure-Aware Fine-Tuning of LLMs for AMR-to-Text Generation [50.277959544420455]
  SAFT is a structure-aware fine-tuning approach that injects graph topology into pretrained language models.<n>We compute direction-sensitive positional encodings from the magnetic Laplacian of transformed AMRs.<n> SAFT sets a new state-of-the-art on AMR 3.0 with a 3.5 BLEU improvement over baselines.
  arXiv  Detail & Related papers  (2025-07-15T18:12:57Z)
• Vector Ontologies as an LLM world view extraction method [0.0]
  Large Language Models (LLMs) possess intricate internal representations of the world, yet these structures are notoriously difficult to interpret or repurpose beyond the original prediction task.<n>A vector ontology defines a domain-specific vector space spanned by ontologically meaningful dimensions, allowing geometric analysis of concepts and relationships within a domain.<n>Using GPT-4o-mini, we extract genre representations through multiple natural language prompts and analyze the consistency of these projections across linguistic variations and their alignment with ground-truth data.
  arXiv  Detail & Related papers  (2025-06-16T08:49:21Z)
• A polar coordinate system represents syntax in large language models [12.244752597245645]
  syntactic trees may also be effectively represented in the activations of large language models.<n>We introduce a 'Polar Probe' trained to read syntactic relations from both the distance and the direction between word embeddings.<n>Our approach reveals three main findings. First, our Polar Probe successfully recovers the type and direction of syntactic relations, and substantially outperforms the Structural Probe by nearly two folds.
  arXiv  Detail & Related papers  (2024-12-07T07:37:20Z)
• Large Language Models are Interpretable Learners [53.56735770834617]
  In this paper, we show a combination of Large Language Models (LLMs) and symbolic programs can bridge the gap between expressiveness and interpretability. The pretrained LLM with natural language prompts provides a massive set of interpretable modules that can transform raw input into natural language concepts. As the knowledge learned by LSP is a combination of natural language descriptions and symbolic rules, it is easily transferable to humans (interpretable) and other LLMs.
  arXiv  Detail & Related papers  (2024-06-25T02:18:15Z)
• Potential and Limitations of LLMs in Capturing Structured Semantics: A Case Study on SRL [78.80673954827773]
  Large Language Models (LLMs) play a crucial role in capturing structured semantics to enhance language understanding, improve interpretability, and reduce bias. We propose using Semantic Role Labeling (SRL) as a fundamental task to explore LLMs' ability to extract structured semantics. We find interesting potential: LLMs can indeed capture semantic structures, and scaling-up doesn't always mirror potential. We are surprised to discover that significant overlap in the errors is made by both LLMs and untrained humans, accounting for almost 30% of all errors.
  arXiv  Detail & Related papers  (2024-05-10T11:44:05Z)
• How Proficient Are Large Language Models in Formal Languages? An In-Depth Insight for Knowledge Base Question Answering [52.86931192259096]
  Knowledge Base Question Answering (KBQA) aims to answer natural language questions based on facts in knowledge bases. Recent works leverage the capabilities of large language models (LLMs) for logical form generation to improve performance.
  arXiv  Detail & Related papers  (2024-01-11T09:27:50Z)
• 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)

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.