mlsanityData to 14 Oct 2026

DirectProbe: Studying Representations without Classifiers

arXiv 2104.05904 ↗First posted 13 Apr 2021cs.CLcited by 25

Not on the map. Why?

Abstract (arXiv)

Understanding how linguistic structures are encoded in contextualized embedding could help explain their impressive performance across NLP@. Existing approaches for probing them usually call for training classifiers and use the accuracy, mutual information, or complexity as a proxy for the representation's goodness. In this work, we argue that doing so can be unreliable because different representations may need different classifiers. We develop a heuristic, DirectProbe, that directly studies the geometry of a representation by building upon the notion of a version space for a task. Experiments with several linguistic tasks and contextualized embeddings show that, even without training classifiers, DirectProbe can shine light into how an embedding space represents labels, and also anticipate classifier performance for the representation.

Key facts. Cited by 25 papers in the map (data through 14 Oct 2026). Builds on 43 of its 61 references that are in the map. First posted on arXiv 13 Apr 2021.

Builds on

43 of 61 references.

References · 43

  1. 2018BERT: Pre-training of Deep Bidirectional Transformers for Language Understandingcited by 40,399
  2. 2014Adam: A Method for Stochastic Optimizationcited by 37,240
  3. 2019RoBERTa: A Robustly Optimized BERT Pretraining Approachcited by 13,372
  4. 2018Deep contextualized word representationscited by 2,290
  5. 2019HuggingFace's Transformers: State-of-the-art Natural Language Processingcited by 2,218
  6. 2016Understanding intermediate layers using linear classifier probescited by 1,353
  7. 2019BERT Rediscovers the Classical NLP Pipelinecited by 1,059
  8. 2019Analyzing Multi-Head Self-Attention: Specialized Heads Do the Heavy Lifting, the Rest Can Be Prunedcited by 856
  9. 2019Are Sixteen Heads Really Better than One?cited by 816
  10. 2019How Contextual are Contextualized Word Representations? Comparing the Geometry of BERT, ELMo, and GPT-2 Embeddingscited by 784
  11. 2019Designing and Interpreting Probes with Control Taskscited by 640
  12. 2018What you can cram into a single vector: Probing sentence embeddings for linguistic propertiescited by 457
31 more
  1. 2019Linguistic Knowledge and Transferability of Contextual Representationscited by 383
  2. 2019Revealing the Dark Secrets of BERTcited by 301
  3. 2019Negated and Misprimed Probes for Pretrained Language Models: Birds Can Talk, But Cannot Flycited by 243
  4. 2018Targeted Syntactic Evaluation of Language Modelscited by 242
  5. 2020Information-Theoretic Probing with Minimum Description Lengthcited by 232
  6. 2020Information-Theoretic Probing for Linguistic Structurecited by 207
  7. 2019To Tune or Not to Tune? Adapting Pretrained Representations to Diverse Taskscited by 167
  8. 2019oLMpics -- On what Language Model Pre-training Capturescited by 161
  9. 2020Unsupervised Domain Clusters in Pretrained Language Modelscited by 160
  10. 2019Do NLP Models Know Numbers? Probing Numeracy in Embeddingscited by 159
  11. 2020Probing the Probing Paradigm: Does Probing Accuracy Entail Task Relevance?cited by 151
  12. 2020Intermediate-Task Transfer Learning with Pretrained Models for Natural Language Understanding: When and Why Does It Work?cited by 127
  13. 2020What Happens To BERT Embeddings During Fine-tuning?cited by 113
  14. 2020Perturbed Masking: Parameter-free Probing for Analyzing and Interpreting BERTcited by 109
  15. 2019Classification and Clustering of Arguments with Contextualized Word Embeddingscited by 57
  16. 2020Pareto Probing: Trading Off Accuracy for Complexitycited by 55
  17. 2020Predicting Performance for Natural Language Processing Taskscited by 49
  18. 2019Probing What Different NLP Tasks Teach Machines about Function Word Comprehensioncited by 44
  19. 2020Probing Linguistic Systematicitycited by 37
  20. 2020Asking without Telling: Exploring Latent Ontologies in Contextual Representationscited by 34
  21. 2020Do Neural Language Models Show Preferences for Syntactic Formalisms?cited by 30
  22. 2018Evaluation of sentence embeddings in downstream and linguistic probing taskscited by 28
  23. 2017Semantic Structure and Interpretability of Word Embeddingscited by 21
  24. 2018Comprehensive Supersense Disambiguation of English Prepositions and Possessivescited by 20
  25. 2020An information theoretic view on selecting linguistic probescited by 19
  26. 2020Evaluating representations by the complexity of learning low-loss predictorscited by 16
  27. 2019Probing for Semantic Classes: Diagnosing the Meaning Content of Word Embeddingscited by 15
  28. 2017How to evaluate word embeddings? On importance of data efficiency and simple supervised taskscited by 12
  29. 2019Pitfalls in the Evaluation of Sentence Embeddingscited by 9
  30. 2019Transfer Fine-Tuning: A BERT Case Studycited by 8
  31. 2017Adposition and Case Supersenses v2.6: Guidelines for Englishcited by 4

Built on by

Cited by year (2026 to 14 Oct)

Citing papers per year: 2021: 3; 2022: 3; 2023: 7; 2024: 2; 2025: 8; 2026: 2.

25 papers.

Sort:

By area:

  1. 2022CREPE: Can Vision-Language Foundation Models Reason Compositionally?cited by 181
  2. 2021A Closer Look at How Fine-tuning Changes BERTcited by 42
  3. 2021How Does Fine-tuning Affect the Geometry of Embedding Space: A Case Study on Isotropycited by 23
  4. 2023The Life Cycle of Knowledge in Big Language Models: A Surveycited by 23
  5. 2023Probing Graph Representationscited by 10
  6. 2021Putting Words in BERT's Mouth: Navigating Contextualized Vector Spaces with Pseudowordscited by 9
  7. 2022Probing via Promptingcited by 7
  8. 2024Evaluating Contextualized Representations of (Spanish) Ambiguous Words: A New Lexical Resource and Empirical Analysiscited by 5
  9. 2022On the data requirements of probingcited by 4
  10. 2024A Critical Study of What Code-LLMs (Do Not) Learncited by 4
  11. 2023CUE: An Uncertainty Interpretation Framework for Text Classifiers Built on Pre-Trained Language Modelscited by 3
  12. 2025On Entity Identification in Language Modelscited by 3
13 more
  1. 2025Towards Transparent AI: A Survey on Explainable Language Modelscited by 3
  2. 2023An Empirical Study on the Transferability of Transformer Modules in Parameter-Efficient Fine-Tuningcited by 2
  3. 2023Analyzing Text Representations by Measuring Task Alignmentcited by 2
  4. 2023Is Probing All You Need? Indicator Tasks as an Alternative to Probing Embedding Spacescited by 2
  5. 2025CAST: Compositional Analysis via Spectral Tracking for Understanding Transformer Layer Functionscited by 2
  6. 2026On the Geometric Structure of Layer Updates in Deep Language Modelscited by 2
  7. 2023Probing Representations for Document-level Event Extractioncited by 1
  8. 2025Investigating the Representation of Backchannels and Fillers in Fine-tuned Language Modelscited by 1
  9. 2025When Annotators Disagree, Topology Explains: Mapper, a Topological Tool for Exploring Text Embedding Geometry and Ambiguitycited by 1
  10. 2026Towards Understanding What State Space Models Learn About Codecited by 1
  11. 2025Domain Pre-training Impact on Representationscited by 0
  12. 2025On the Performance of Concept Probing: The Influence of the Data (Extended Version)cited by 0
  13. 2025Concept Probing: Where to Find Human-Defined Concepts (Extended Version)cited by 0