mlsanityData to 14 Oct 2026

Machine Reading, Fast and Slow: When Do Models "Understand" Language?

arXiv 2209.07430 ↗First posted 15 Sep 2022cs.CLcited by 13

Not on the map. Why?

Abstract (arXiv)

Two of the most fundamental challenges in Natural Language Understanding (NLU) at present are: (a) how to establish whether deep learning-based models score highly on NLU benchmarks for the 'right' reasons; and (b) to understand what those reasons would even be. We investigate the behavior of reading comprehension models with respect to two linguistic 'skills': coreference resolution and comparison. We propose a definition for the reasoning steps expected from a system that would be 'reading slowly', and compare that with the behavior of five models of the BERT family of various sizes, observed through saliency scores and counterfactual explanations. We find that for comparison (but not coreference) the systems based on larger encoders are more likely to rely on the 'right' information, but even they struggle with generalization, suggesting that they still learn specific lexical patterns rather than the general principles of comparison.

Key facts. Cited by 13 papers in the map (data through 14 Oct 2026). Builds on 55 of its 88 references that are in the map. First posted on arXiv 15 Sep 2022.

Builds on

55 of 88 references.

References · 55

  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. 2019Sentence-BERT: Sentence Embeddings using Siamese BERT-Networkscited by 9,170
  5. 2016"Why Should I Trust You?": Explaining the Predictions of Any Classifiercited by 5,070
  6. 2018GLUE: A Multi-Task Benchmark and Analysis Platform for Natural Language Understandingcited by 4,899
  7. 2013Efficient Estimation of Word Representations in Vector Spacecited by 4,375
  8. 2016SQuAD: 100,000+ Questions for Machine Comprehension of Textcited by 4,148
  9. 2018HotpotQA: A Dataset for Diverse, Explainable Multi-hop Question Answeringcited by 3,754
  10. 2017Axiomatic Attribution for Deep Networkscited by 3,384
  11. 2019SuperGLUE: A Stickier Benchmark for General-Purpose Language Understanding Systemscited by 1,729
  12. 2020Constructing A Multi-hop QA Dataset for Comprehensive Evaluation of Reasoning Stepscited by 1,282
38 more
  1. 2014Convolutional Neural Networks for Sentence Classificationcited by 1,127
  2. 2019Right for the Wrong Reasons: Diagnosing Syntactic Heuristics in Natural Language Inferencecited by 876
  3. 2019Is BERT Really Robust? A Strong Baseline for Natural Language Attack on Text Classification and Entailmentcited by 671
  4. 2019MathQA: Towards Interpretable Math Word Problem Solving with Operation-Based Formalismscited by 662
  5. 2017Adversarial Examples for Evaluating Reading Comprehension Systemscited by 571
  6. 2020Captum: A unified and generic model interpretability library for PyTorchcited by 547
  7. 2019ERASER: A Benchmark to Evaluate Rationalized NLP Modelscited by 530
  8. 2019Learning the Difference that Makes a Difference with Counterfactually-Augmented Datacited by 439
  9. 2020Pretrained Transformers Improve Out-of-Distribution Robustnesscited by 294
  10. 2018Analysis Methods in Neural Language Processing: A Surveycited by 277
  11. 2019Negated and Misprimed Probes for Pretrained Language Models: Birds Can Talk, But Cannot Flycited by 243
  12. 2016Understanding Neural Networks through Representation Erasurecited by 224
  13. 2020A Diagnostic Study of Explainability Techniques for Text Classificationcited by 159
  14. 2020Explaining Black Box Predictions and Unveiling Data Artifacts through Influence Functionscited by 144
  15. 2019Well-Read Students Learn Better: On the Importance of Pre-training Compact Modelscited by 120
  16. 2020How Can We Accelerate Progress Towards Human-like Linguistic Generalization?cited by 114
  17. 2019Quoref: A Reading Comprehension Dataset with Questions Requiring Coreferential Reasoningcited by 100
  18. 2021Competency Problems: On Finding and Removing Artifacts in Language Datacited by 91
  19. 2021QA Dataset Explosion: A Taxonomy of NLP Resources for Question Answering and Reading Comprehensioncited by 91
  20. 2018How Much Reading Does Reading Comprehension Require? A Critical Investigation of Popular Benchmarkscited by 88
  21. 2019Compositional Questions Do Not Necessitate Multi-hop Reasoningcited by 88
  22. 2020Look at the First Sentence: Position Bias in Question Answeringcited by 84
  23. 2016A Thorough Examination of the CNN/Daily Mail Reading Comprehension Taskcited by 72
  24. 2021Generalization in NLI: Ways (Not) To Go Beyond Simple Heuristicscited by 67
  25. 2016Deep Reinforcement Learning for Mention-Ranking Coreference Modelscited by 63
  26. 2018What Makes Reading Comprehension Questions Easier?cited by 60
  27. 2020QED: A Framework and Dataset for Explanations in Question Answeringcited by 55
  28. 2019Understanding Dataset Design Choices for Multi-hop Reasoningcited by 50
  29. 2021GLocalX -- From Local to Global Explanations of Black Box AI Modelscited by 48
  30. 2019Assessing the Benchmarking Capacity of Machine Reading Comprehension Datasetscited by 46
  31. 2020What do Models Learn from Question Answering Datasets?cited by 45
  32. 2021Why AI is Harder Than We Thinkcited by 41
  33. 2020jiant: A Software Toolkit for Research on General-Purpose Text Understanding Modelscited by 39
  34. 2020To Test Machine Comprehension, Start by Defining Comprehensioncited by 31
  35. 2021Evaluating the Faithfulness of Importance Measures in NLP by Recursively Masking Allegedly Important Tokens and Retrainingcited by 31
  36. 2021Connecting Attributions and QA Model Behavior on Realistic Counterfactualscited by 20
  37. 2021How Does Counterfactually Augmented Data Impact Models for Social Computing Constructs?cited by 19
  38. 2020A Framework for Evaluation of Machine Reading Comprehension Gold Standardscited by 18

50 most cited of 55 All 55 (JSON)

Built on by

Cited by year (2026 to 14 Oct)

Citing papers per year: 2022: 2; 2023: 4; 2024: 4; 2025: 2; 2026: 1.

13 papers.

Sort:

By area:

  1. 2022The Debate Over Understanding in AI's Large Language Modelscited by 142
  2. 2023Position: Key Claims in LLM Research Have a Long Tail of Footnotescited by 20
  3. 2023Separating form and meaning: Using self-consistency to quantify task understanding across multiple sensescited by 15
  4. 2024CLEAR: Can Language Models Really Understand Causal Graphs?cited by 14
  5. 2023Being Right for Whose Right Reasons?cited by 12
  6. 2024Natural Language Processing RELIES on Linguisticscited by 9
  7. 2024From Form(s) to Meaning: Probing the Semantic Depths of Language Models Using Multisense Consistencycited by 7
  8. 2023The Two Word Test: A Semantic Benchmark for Large Language Modelscited by 5
  9. 2022Implicit causality in GPT-2: a case studycited by 2
  10. 2025Research Community Perspectives on "Intelligence" and Large Language Modelscited by 2
  11. 2025Knowing the Facts but Choosing the Shortcut: Understanding How Large Language Models Compare Entitiescited by 2
  12. 2026The Overlooked Repetitive Lengthening Form in Sentiment Analysiscited by 1
1 more
  1. 2024FAC2E: Better Understanding Large Language Model Capabilities by Dissociating Language and Cognitioncited by 0

Similar, not cited

Close in wording, no citation either way

  1. 2022A Survey on Measuring and Mitigating Reasoning Shortcuts in Machine Reading Comprehension
  2. 2023Large Language Models Are Not Strong Abstract Reasoners
  3. 2024Can large language models understand uncommon meanings of common words?
  4. 2022The KITMUS Test: Evaluating Knowledge Integration from Multiple Sources in Natural Language Understanding Systems
  5. 2025Interpreting Language Models Through Concept Descriptions: A Survey
  6. 2024Interpretability of Language Models via Task Spaces
  7. 2024MMLU-SR: A Benchmark for Stress-Testing Reasoning Capability of Large Language Models
  8. 2025Demystifying the Roles of LLM Layers in Retrieval, Knowledge, and Reasoning