mlsanityData to 7 Oct 2026

Visualizing and Measuring the Geometry of BERT

arXiv 1906.02715 ↗First posted 6 Jun 2019cs.LG cs.CL stat.MLcited by 215

Not on the map. Why?

Abstract (arXiv)

Transformer architectures show significant promise for natural language processing. Given that a single pretrained model can be fine-tuned to perform well on many different tasks, these networks appear to extract generally useful linguistic features. A natural question is how such networks represent this information internally. This paper describes qualitative and quantitative investigations of one particularly effective model, BERT. At a high level, linguistic features seem to be represented in separate semantic and syntactic subspaces. We find evidence of a fine-grained geometric representation of word senses. We also present empirical descriptions of syntactic representations in both attention matrices and individual word embeddings, as well as a mathematical argument to explain the geometry of these representations.

Key facts. Cited by 215 papers in the map (data through 7 Oct 2026). First posted on arXiv 6 Jun 2019.

Builds on

No reference list yet.

Built on by

Cited by year (2026 to 7 Oct)

Citing papers per year: 2021: 53; 2022: 31; 2023: 45; 2024: 36; 2025: 28; 2026: 22.

215 papers.

Sort:

By area:

  1. 2023The Linear Representation Hypothesis and the Geometry of Large Language Modelscited by 679
  2. 2023Dissecting Recall of Factual Associations in Auto-Regressive Language Modelscited by 512
  3. 2023Finding Neurons in a Haystack: Case Studies with Sparse Probingcited by 328
  4. 2021Large Pre-trained Language Models Contain Human-like Biases of What is Right and Wrong to Docited by 178
  5. 2021Post-hoc Interpretability for Neural NLP: A Surveycited by 155
  6. 2024The Geometry of Categorical and Hierarchical Concepts in Large Language Modelscited by 151
  7. 2021All Bark and No Bite: Rogue Dimensions in Transformer Language Models Obscure Representational Qualitycited by 150
  8. 2022Towards Faithful Model Explanation in NLP: A Surveycited by 129
  9. 2021COCO-LM: Correcting and Contrasting Text Sequences for Language Model Pretrainingcited by 126
  10. 2022Toward Transparent AI: A Survey on Interpreting the Inner Structures of Deep Neural Networkscited by 126
  11. 2021Can Language Models Encode Perceptual Structure Without Grounding? A Case Study in Colorcited by 118
  12. 2021Transformer visualization via dictionary learning: contextualized embedding as a linear superposition of transformer factorscited by 106
38 more
  1. 2021BBAM: Bounding Box Attribution Map for Weakly Supervised Semantic and Instance Segmentationcited by 104
  2. 2023Interpretability at Scale: Identifying Causal Mechanisms in Alpacacited by 102
  3. 2021Pose Recognition with Cascade Transformerscited by 99
  4. 2023Language Model Behavior: A Comprehensive Surveycited by 92
  5. 2024On the Origins of Linear Representations in Large Language Modelscited by 76
  6. 2021Mitigating Political Bias in Language Models Through Reinforced Calibrationcited by 73
  7. 2021Superbizarre Is Not Superb: Derivational Morphology Improves BERT's Interpretation of Complex Wordscited by 63
  8. 2021The Spotlight: A General Method for Discovering Systematic Errors in Deep Learning Modelscited by 63
  9. 2022Discovering Latent Concepts Learned in BERTcited by 56
  10. 2021IsoScore: Measuring the Uniformity of Embedding Space Utilizationcited by 52
  11. 2021Probing BERT in Hyperbolic Spacescited by 50
  12. 2021Incorporating Residual and Normalization Layers into Analysis of Masked Language Modelscited by 50
  13. 2023AttentionViz: A Global View of Transformer Attentioncited by 50
  14. 2021How Should Pre-Trained Language Models Be Fine-Tuned Towards Adversarial Robustness?cited by 49
  15. 2022Problems with Cosine as a Measure of Embedding Similarity for High Frequency Wordscited by 49
  16. 2023Do Transformers Parse while Predicting the Masked Word?cited by 48
  17. 2023Are Transformers with One Layer Self-Attention Using Low-Rank Weight Matrices Universal Approximators?cited by 46
  18. 2023Theory of Mind in Large Language Models: Examining Performance of 11 State-of-the-Art models vs. Children Aged 7-10 on Advanced Testscited by 46
  19. 2022GlobEnc: Quantifying Global Token Attribution by Incorporating the Whole Encoder Layer in Transformerscited by 44
  20. 2023Quantifying Context Mixing in Transformerscited by 43
  21. 2026When Models Manipulate Manifolds: The Geometry of a Counting Taskcited by 43
  22. 2022SemAttack: Natural Textual Attacks via Different Semantic Spacescited by 42
  23. 2021Let's Play Mono-Poly: BERT Can Reveal Words' Polysemy Level and Partitionability into Sensescited by 39
  24. 2021The Low-Dimensional Linear Geometry of Contextualized Word Representationscited by 39
  25. 2021What if This Modified That? Syntactic Interventions via Counterfactual Embeddingscited by 39
  26. 2022Analyzing Encoded Concepts in Transformer Language Modelscited by 37
  27. 2021Distilling Linguistic Context for Language Model Compressioncited by 33
  28. 2021A Non-Linear Structural Probecited by 31
  29. 2022Proton: Probing Schema Linking Information from Pre-trained Language Models for Text-to-SQL Parsingcited by 30
  30. 2023A Fast Optimization View: Reformulating Single Layer Attention in LLM Based on Tensor and SVM Trick, and Solving It in Matrix Multiplication Timecited by 30
  31. 2021Exploring the Role of BERT Token Representations to Explain Sentence Probing Resultscited by 29
  32. 2021A Cluster-based Approach for Improving Isotropy in Contextual Embedding Spacecited by 27
  33. 2024Mechanistic?cited by 25
  34. 2024Hyperbolic Fine-Tuning for Large Language Modelscited by 24
  35. 2021Learning to Remove: Towards Isotropic Pre-trained BERT Embeddingcited by 23
  36. 2021How Does Fine-tuning Affect the Geometry of Embedding Space: A Case Study on Isotropycited by 23
  37. 2022STNDT: Modeling Neural Population Activity with a Spatiotemporal Transformercited by 23
  38. 2023GradientCoin: A Peer-to-Peer Decentralized Large Language Modelscited by 23

50 most cited of 215 All 215 (JSON)