mlsanityData to 7 Oct 2026

Impact of Pretraining Term Frequencies on Few-Shot Reasoning

arXiv 2202.07206 ↗First posted 15 Feb 2022cs.CL cs.LGcited by 142

On the map: LLM reasoning → Reasoning benchmarks and evaluation

Abstract (arXiv)

Pretrained Language Models (LMs) have demonstrated ability to perform numerical reasoning by extrapolating from a few examples in few-shot settings. However, the extent to which this extrapolation relies on robust reasoning is unclear. In this paper, we investigate how well these models reason with terms that are less frequent in the pretraining data. In particular, we examine the correlations between the model performance on test instances and the frequency of terms from those instances in the pretraining data. We measure the strength of this correlation for a number of GPT-based language models (pretrained on the Pile dataset) on various numerical deduction tasks (e.g., arithmetic and unit conversion). Our results consistently demonstrate that models are more accurate on instances whose terms are more prevalent, in some cases above 70% (absolute) more accurate on the top 10% frequent terms in comparison to the bottom 10%. Overall, although LMs exhibit strong performance at few-shot numerical reasoning tasks, our results raise the question of how much models actually generalize beyond pretraining data, and we encourage researchers to take the pretraining data into account when interpreting evaluation results.

Key facts. Cited by 142 papers in the map (data through 7 Oct 2026). Placed in Reasoning benchmarks and evaluation, one of 17 directions in LLM reasoning. Builds on 40 of its 55 references that are in the map. First posted on arXiv 15 Feb 2022.

Builds on

40 of 55 references.

References · 40

  1. 2020Language Models are Few-Shot Learnerscited by 32,199
  2. 2019Exploring the Limits of Transfer Learning with a Unified Text-to-Text Transformercited by 14,491
  3. 2021On the Opportunities and Risks of Foundation Modelscited by 3,961
  4. 2020The Pile: An 800GB Dataset of Diverse Text for Language Modelingcited by 2,346
  5. 2019Language Models as Knowledge Bases?cited by 1,922
  6. 2018Datasheets for Datasetscited by 1,371
  7. 2020RealToxicityPrompts: Evaluating Neural Toxic Degeneration in Language Modelscited by 1,235
  8. 2019Right for the Wrong Reasons: Diagnosing Syntactic Heuristics in Natural Language Inferencecited by 864
  9. 2021Are NLP Models really able to Solve Simple Math Word Problems?cited by 861
  10. 2018The Secret Sharer: Evaluating and Testing Unintended Memorization in Neural Networkscited by 847
  11. 2018Annotation Artifacts in Natural Language Inference Datacited by 656
  12. 2017Adversarial Examples for Evaluating Reading Comprehension Systemscited by 568
28 more
  1. 2020What Neural Networks Memorize and Why: Discovering the Long Tail via Influence Estimationcited by 492
  2. 2021Documenting Large Webtext Corpora: A Case Study on the Colossal Clean Crawled Corpuscited by 481
  3. 2019Does Learning Require Memorization? A Short Tale about a Long Tailcited by 452
  4. 2018Gender Bias in Coreference Resolutioncited by 438
  5. 2019Analysing Mathematical Reasoning Abilities of Neural Modelscited by 331
  6. 2018Hypothesis Only Baselines in Natural Language Inferencecited by 291
  7. 2020AutoPrompt: Eliciting Knowledge from Language Models with Automatically Generated Promptscited by 246
  8. 2019Commonsense Knowledge Mining from Pretrained Modelscited by 189
  9. 2021Masked Language Modeling and the Distributional Hypothesis: Order Word Matters Pre-training for Littlecited by 176
  10. 2019Do NLP Models Know Numbers? Probing Numeracy in Embeddingscited by 157
  11. 2021Counterfactual Memorization in Neural Language Modelscited by 155
  12. 2020Question and Answer Test-Train Overlap in Open-Domain Question Answering Datasetscited by 137
  13. 2020Injecting Numerical Reasoning Skills into Language Modelscited by 133
  14. 2019End-to-End Bias Mitigation by Modelling Biases in Corporacited by 127
  15. 2020Birds have four legs?! NumerSense: Probing Numerical Commonsense Knowledge of Pre-trained Language Modelscited by 110
  16. 2021Competency Problems: On Finding and Removing Artifacts in Language Datacited by 91
  17. 2021What's in the Box? A Preliminary Analysis of Undesirable Content in the Common Crawl Corpuscited by 79
  18. 2016A Thorough Examination of the CNN/Daily Mail Reading Comprehension Taskcited by 71
  19. 2020We Need to Talk About Random Splitscited by 70
  20. 2021Memorization vs. Generalization: Quantifying Data Leakage in NLP Performance Evaluationcited by 69
  21. 2018What Makes Reading Comprehension Questions Easier?cited by 59
  22. 2021Frequency Effects on Syntactic Rule Learning in Transformerscited by 59
  23. 2019What's in a Name? Reducing Bias in Bios without Access to Protected Attributescited by 51
  24. 2018Numeracy for Language Models: Evaluating and Improving their Ability to Predict Numberscited by 42
  25. 2020Temporal Common Sense Acquisition with Minimal Supervisioncited by 38
  26. 2021Evaluating Entity Disambiguation and the Role of Popularity in Retrieval-Based NLPcited by 30
  27. 2020Poison Attacks against Text Datasets with Conditional Adversarially Regularized Autoencodercited by 27
  28. 2020Know thy corpus! Robust methods for digital curation of Web corporacited by 6

Built on by

Cited by year (2026 to 7 Oct)

Citing papers per year: 2022: 47; 2023: 45; 2024: 22; 2025: 22; 2026: 6.

142 papers.

Sort:

By area:

  1. 2022Self-Instruct: Aligning Language Models with Self-Generated Instructionscited by 2,674
  2. 2022Emergent Abilities of Large Language Modelscited by 2,483
  3. 2023Pythia: A Suite for Analyzing Large Language Models Across Training and Scalingcited by 1,734
  4. 2022Rethinking the Role of Demonstrations: What Makes In-Context Learning Work?cited by 1,413
  5. 2022When Not to Trust Language Models: Investigating Effectiveness of Parametric and Non-Parametric Memoriescited by 1,103
  6. 2022Galactica: A Large Language Model for Sciencecited by 736
  7. 2022GPT-NeoX-20B: An Open-Source Autoregressive Language Modelcited by 711
  8. 2022What Can Transformers Learn In-Context? A Case Study of Simple Function Classescited by 700
  9. 2022Towards Reasoning in Large Language Models: A Surveycited by 593
  10. 2024GSM-Symbolic: Understanding the Limitations of Mathematical Reasoning in Large Language Modelscited by 533
  11. 2022Large Language Models Struggle to Learn Long-Tail Knowledgecited by 452
  12. 2022Language Models Are Greedy Reasoners: A Systematic Formal Analysis of Chain-of-Thoughtcited by 334
38 more
  1. 2023Data Selection for Language Models via Importance Resamplingcited by 322
  2. 2022Large Language Models Are Reasoning Teacherscited by 318
  3. 2022Data Distributional Properties Drive Emergent In-Context Learning in Transformerscited by 303
  4. 2023Transformers as Statisticians: Provable In-Context Learning with In-Context Algorithm Selectioncited by 291
  5. 2022Data Contamination: From Memorization to Exploitationcited by 234
  6. 2022Towards Understanding Chain-of-Thought Prompting: An Empirical Study of What Matterscited by 221
  7. 2022Demystifying Prompts in Language Models via Perplexity Estimationcited by 215
  8. 2022Exploring Length Generalization in Large Language Modelscited by 204
  9. 2023Time Travel in LLMs: Tracing Data Contamination in Large Language Modelscited by 175
  10. 2022Language models show human-like content effects on reasoning taskscited by 171
  11. 2023Birth of a Transformer: A Memory Viewpointcited by 154
  12. 2023Supervised Pretraining Can Learn In-Context Reinforcement Learningcited by 137
  13. 2023How well do Large Language Models perform in Arithmetic tasks?cited by 130
  14. 2022Text and Patterns: For Effective Chain of Thought, It Takes Two to Tangocited by 128
  15. 2022GrIPS: Gradient-free, Edit-based Instruction Search for Prompting Large Language Modelscited by 127
  16. 2022A Survey of Deep Learning for Mathematical Reasoningcited by 124
  17. 2023Teaching Arithmetic to Small Transformerscited by 121
  18. 2023ReviewerGPT? An Exploratory Study on Using Large Language Models for Paper Reviewingcited by 118
  19. 2022Teaching Algorithmic Reasoning via In-context Learningcited by 111
  20. 2022State-of-the-art generalisation research in NLP: A taxonomy and reviewcited by 97
  21. 2023Mitigating Label Biases for In-context Learningcited by 88
  22. 2022Mind's Eye: Grounded Language Model Reasoning through Simulationcited by 83
  23. 2022Structured Prompting: Scaling In-Context Learning to 1,000 Examplescited by 81
  24. 2023In-Context Learning through the Bayesian Prismcited by 78
  25. 2023How Do Transformers Learn In-Context Beyond Simple Functions? A Case Study on Learning with Representationscited by 72
  26. 2023Generative Speech Recognition Error Correction with Large Language Models and Task-Activating Promptingcited by 71
  27. 2022Z-ICL: Zero-Shot In-Context Learning with Pseudo-Demonstrationscited by 70
  28. 2022A Causal Framework to Quantify the Robustness of Mathematical Reasoning with Language Modelscited by 68
  29. 2022Rethinking the Role of Scale for In-Context Learning: An Interpretability-based Case Study at 66 Billion Scalecited by 64
  30. 2023Understanding In-Context Learning in Transformers and LLMs by Learning to Learn Discrete Functionscited by 62
  31. 2023In-context Example Selection with Influencescited by 60
  32. 2023GPT4AIGChip: Towards Next-Generation AI Accelerator Design Automation via Large Language Modelscited by 59
  33. 2022Prototypical Calibration for Few-shot Learning of Language Modelscited by 58
  34. 2022Measuring Causal Effects of Data Statistics on Language Model's `Factual' Predictionscited by 57
  35. 2022Can Large Language Models Truly Understand Prompts? A Case Study with Negated Promptscited by 56
  36. 2025FineVision: Open Data Is All You Needcited by 51
  37. 2023In-Context Learning Learns Label Relationships but Is Not Conventional Learningcited by 48
  38. 2023Characterizing Mechanisms for Factual Recall in Language Modelscited by 47

50 most cited of 142 All 142 (JSON)

Similar, not cited

Close in wording, no citation either way

  1. 2024Physics of Language Models: Part 2.2, How to Learn From Mistakes on Grade-School Math Problems
  2. 2024Procedural Knowledge in Pretraining Drives Reasoning in Large Language Models
  3. 2022Large Language Models are Zero-Shot Reasoners
  4. 2025Reasoning Capabilities and Invariability of Large Language Models
  5. 2024What Do Learning Dynamics Reveal About Generalization in LLM Reasoning?
  6. 2023Specializing Smaller Language Models towards Multi-Step Reasoning
  7. 2025Mathematical Reasoning in Large Language Models: Assessing Logical and Arithmetic Errors across Wide Numerical Ranges
  8. 2025Front-Loading Reasoning: The Synergy between Pretraining and Post-Training Data