mlsanityData to 14 Oct 2026

Emergent Symbolic Mechanisms Support Abstract Reasoning in Large Language Models

arXiv 2502.20332 ↗First posted 27 Feb 2025cs.CL cs.AIcited by 31

On the map: LLM reasoning → Reasoning benchmarks and evaluation

Abstract (arXiv)

Many recent studies have found evidence for emergent reasoning capabilities in large language models (LLMs), but debate persists concerning the robustness of these capabilities, and the extent to which they depend on structured reasoning mechanisms. To shed light on these issues, we study the internal mechanisms that support abstract reasoning in LLMs. We identify an emergent symbolic architecture that implements abstract reasoning via a series of three computations. In early layers, symbol abstraction heads convert input tokens to abstract variables based on the relations between those tokens. In intermediate layers, symbolic induction heads perform sequence induction over these abstract variables. Finally, in later layers, retrieval heads predict the next token by retrieving the value associated with the predicted abstract variable. These results point toward a resolution of the longstanding debate between symbolic and neural network approaches, suggesting that emergent reasoning in neural networks depends on the emergence of symbolic mechanisms.

Key facts. Cited by 31 papers in the map (data through 14 Oct 2026). Placed in Reasoning benchmarks and evaluation, one of 17 directions in LLM reasoning. Builds on 31 of its 52 references that are in the map. First posted on arXiv 27 Feb 2025.

Builds on

31 of 52 references.

Closest ancestors · 6

Same direction, newest first

  1. 2024Evaluating the Robustness of Analogical Reasoning in Large Language Models21 Nov
  2. 2024Evidence from counterfactual tasks supports emergent analogical reasoning in large language models14 Apr
  3. 2023Reasoning or Reciting? Exploring the Capabilities and Limitations of Language Models Through Counterfactual Tasks5 Jul
  4. 2023From Word Models to World Models: Translating from Natural Language to the Probabilistic Language of Thought22 Jun
  5. 2022Emergent Analogical Reasoning in Large Language Models19 Dec
  6. 2022Language models show human-like content effects on reasoning tasks14 Jul

Other references · 25

  1. 2017Attention Is All You Needcited by 60,361
  2. 2024The Llama 3 Herd of Modelscited by 14,715
  3. 2024Qwen2.5 Technical Reportcited by 4,942
  4. 2022Locating and Editing Factual Associations in GPTcited by 2,691
  5. 2022Emergent Abilities of Large Language Modelscited by 2,512
  6. 2024Gemma 2: Improving Open Language Models at a Practical Sizecited by 1,998
  7. 2022Interpretability in the Wild: a Circuit for Indirect Object Identification in GPT-2 smallcited by 1,165
  8. 2022In-context Learning and Induction Headscited by 964
  9. 2023Faith and Fate: Limits of Transformers on Compositionalitycited by 503
  10. 2023Function Vectors in Large Language Modelscited by 308
  11. 2013Direct and Indirect Effectscited by 280
  12. 2023Large Language Models as General Pattern Machinescited by 197
13 more
  1. 2023Embers of Autoregression: Understanding Large Language Models Through the Problem They are Trained to Solvecited by 178
  2. 2023How do Language Models Bind Entities in Context?cited by 108
  3. 2024In-Context Language Learning: Architectures and Algorithmscited by 89
  4. 2023Evaluating Cognitive Maps and Planning in Large Language Models with CogEvalcited by 77
  5. 2025Which Attention Heads Matter for In-Context Learning?cited by 61
  6. 2024A Mechanistic Analysis of a Transformer Trained on a Symbolic Multi-Step Reasoning Taskcited by 49
  7. 2020Emergent Symbols through Binding in External Memorycited by 47
  8. 2023The Relational Bottleneck as an Inductive Bias for Efficient Abstractioncited by 27
  9. 2022On Neural Architecture Inductive Biases for Relational Taskscited by 26
  10. 2023Abstractors and relational cross-attention: An inductive bias for explicit relational reasoning in Transformerscited by 12
  11. 2025An explainable transformer circuit for compositional generalizationcited by 8
  12. 2025Emergent Symbol-like Number Variables in Artificial Neural Networkscited by 6
  13. 2024Approximation of relation functions and attention mechanismscited by 5

Built on by

Cited by year (2026 to 14 Oct)

Citing papers per year: 2025: 18; 2026: 13.

31 papers.

Sort:

By area:

  1. 2025Language Models use Lookbacks to Track Beliefscited by 39
  2. 2025Emergent Abilities in Large Language Models: A Surveycited by 37
  3. 2025The Dual-Route Model of Inductioncited by 24
  4. 2025Stepwise Guided Policy Optimization: Coloring your Incorrect Reasoning in GRPOcited by 19
  5. 2025Causal Head Gating: A Framework for Interpreting Roles of Attention Heads in Transformerscited by 16
  6. 2025Visual symbolic mechanisms: Emergent symbol processing in vision language modelscited by 13
  7. 2025Decomposing Representation Space into Interpretable Subspaces with Unsupervised Learningcited by 8
  8. 2026Causality neq Invariance: Function and Concept Vectors in LLMscited by 8
  9. 2025Whither symbols in the era of advanced neural networks?cited by 7
  10. 2025LLMs Process Lists With General Filter Headscited by 7
  11. 2025The Mechanistic Emergence of Symbol Grounding in Language Modelscited by 6
  12. 2025Learning to Ponder: Adaptive Reasoning in Latent Spacecited by 5
19 more
  1. 2025In-Context Algebracited by 5
  2. 2025Illusion or Algorithm? Investigating Memorization, Emergence, and Symbolic Processing in In-Context Learningcited by 3
  3. 2026Understanding In-Context Learning for Nonlinear Regression with Transformers: Attention as Featurizercited by 3
  4. 2025A circuit for predicting hierarchical structure in-context in Large Language Modelscited by 1
  5. 2025Understanding the Staged Dynamics of Transformers in Learning Latent Structurecited by 1
  6. 2026Can Large Language Models Generalize Procedures Across Representations?cited by 1
  7. 2026Grokking From Abstraction to Intelligencecited by 1
  8. 2026Mechanisms of Object Localization in Vision-Language Modelscited by 1
  9. 2026Who Says What: Symbolic Trimodal Binding Mechanisms in Audio-Visual LLMscited by 1
  10. 2025A Group Theoretic Analysis of the Symmetries Underlying Base Addition and Their Learnability by Neural Networkscited by 0
  11. 2025Representing Beauty: Towards a Participatory but Objective Latent Aestheticscited by 0
  12. 2025In-Context Learning Without Copyingcited by 0
  13. 2026Emergent Structured Representations Support Flexible In-Context Inference in Large Language Modelscited by 0
  14. 2026How Do Language Models Understand Tables? A Mechanistic Analysis of Cell Locationcited by 0
  15. 2026Binding Visual Features Point by Pointcited by 0
  16. 2026HPP: Hierarchical Programmatic Probing for Long Video Understanding by Decoupling Perception and Reasoningcited by 0
  17. 2026A Pre-Specified Construction-Confirmation Test of Operation-Level Causal Transfer Across Finite Isomorphic Symbolic Domainscited by 0
  18. 2026Logical subspace in LLMscited by 0
  19. 2026In-Context Learning Amplifies a Latent Symbolic Circuitcited by 0

Similar, not cited

Close in wording, no citation either way

  1. 2026Unveiling the Reasoning Process of Large Language Models
  2. 2026The Emergent Symbolic Structure of Artificial Neural Networks
  3. 2025Interpretable by AI Mother Tongue: Native Symbolic Reasoning in Neural Models
  4. 2026Compositional Neuro-Symbolic Reasoning
  5. 2023Empirical Investigation of Neural Symbolic Reasoning Strategies
  6. 2024Reasoning Circuits in Language Models: A Mechanistic Interpretation of Syllogistic Inference
  7. 2026Revealing Algorithmic Deductive Circuits for Logical Reasoning
  8. 2023Large Language Models are In-Context Semantic Reasoners rather than Symbolic Reasoners