mlsanityData to 7 Oct 2026

Describing Differences between Text Distributions with Natural Language

arXiv 2201.12323 ↗First posted 28 Jan 2022cs.CL cs.AI cs.LGcited by 54

Not on the map. Why?

Abstract (arXiv)

How do two distributions of texts differ? Humans are slow at answering this, since discovering patterns might require tediously reading through hundreds of samples. We propose to automatically summarize the differences by "learning a natural language hypothesis": given two distributions D0 and D1, we search for a description that is more often true for D1, e.g., "is military-related." To tackle this problem, we fine-tune GPT-3 to propose descriptions with the prompt: "[samples of D0] + [samples of D1] + the difference between them is___." We then re-rank the descriptions by checking how often they hold on a larger set of samples with a learned verifier. On a benchmark of 54 real-world binary classification tasks, while GPT-3 Curie (13B) only generates a description similar to human annotation 7% of the time, the performance reaches 61% with fine-tuning and re-ranking, and our best system using GPT-3 Davinci (175B) reaches 76%. We apply our system to describe distribution shifts, debug dataset shortcuts, summarize unknown tasks, and label text clusters, and present analyses based on automatically generated descriptions.

Key facts. Cited by 54 papers in the map (data through 7 Oct 2026). Builds on 47 of its 82 references that are in the map. First posted on arXiv 28 Jan 2022.

Builds on

47 of 82 references.

References · 47

  1. 2021Learning Transferable Visual Models From Natural Language Supervisioncited by 33,456
  2. 2020Language Models are Few-Shot Learnerscited by 32,199
  3. 2019Exploring the Limits of Transfer Learning with a Unified Text-to-Text Transformercited by 14,491
  4. 2019RoBERTa: A Robustly Optimized BERT Pretraining Approachcited by 13,266
  5. 2021Training Verifiers to Solve Math Word Problemscited by 9,743
  6. 2019BERTScore: Evaluating Text Generation with BERTcited by 5,462
  7. 2021Prefix-Tuning: Optimizing Continuous Prompts for Generationcited by 3,878
  8. 2021Finetuned Language Models Are Zero-Shot Learnerscited by 3,747
  9. 2016Pointer Sentinel Mixture Modelscited by 2,992
  10. 2017A Broad-Coverage Challenge Corpus for Sentence Understanding through Inferencecited by 2,665
  11. 2015Character-level Convolutional Networks for Text Classificationcited by 2,293
  12. 2019BoolQ: Exploring the Surprising Difficulty of Natural Yes/No Questionscited by 2,140
35 more
  1. 2020The Many Faces of Robustness: A Critical Analysis of Out-of-Distribution Generalizationcited by 1,842
  2. 2015A large annotated corpus for learning natural language inferencecited by 1,837
  3. 2021WebGPT: Browser-assisted question-answering with human feedbackcited by 1,722
  4. 2016Concrete Problems in AI Safetycited by 1,582
  5. 2021Multitask Prompted Training Enables Zero-Shot Task Generalizationcited by 1,462
  6. 2020Making Pre-trained Language Models Better Few-shot Learnerscited by 1,255
  7. 2019Distributionally Robust Neural Networks for Group Shifts: On the Importance of Regularization for Worst-Case Generalizationcited by 1,232
  8. 2021Scaling Language Models: Methods, Analysis & Insights from Training Gophercited by 1,199
  9. 2018Neural Network Acceptability Judgmentscited by 952
  10. 2021Robust fine-tuning of zero-shot modelscited by 785
  11. 2021Cross-Task Generalization via Natural Language Crowdsourcing Instructionscited by 696
  12. 2018Annotation Artifacts in Natural Language Inference Datacited by 656
  13. 2021Learning To Retrieve Prompts for In-Context Learningcited by 641
  14. 2020UnifiedQA: Crossing Format Boundaries With a Single QA Systemcited by 475
  15. 2020TweetEval: Unified Benchmark and Comparative Evaluation for Tweet Classificationcited by 414
  16. 2018Fairness Without Demographics in Repeated Loss Minimizationcited by 363
  17. 2019Benchmarking Zero-shot Text Classification: Datasets, Evaluation and Entailment Approachcited by 299
  18. 2021Annotators with Attitudes: How Annotator Beliefs And Identities Bias Toxic Language Detectioncited by 258
  19. 2015Strategic Classificationcited by 256
  20. 2020AutoPrompt: Eliciting Knowledge from Language Models with Automatically Generated Promptscited by 246
  21. 2018Stress Test Evaluation for Natural Language Inferencecited by 211
  22. 2021Reframing Instructional Prompts to GPTk's Languagecited by 176
  23. 2021Noisy Channel Language Model Prompting for Few-Shot Text Classificationcited by 163
  24. 2020Unsupervised Domain Clusters in Pretrained Language Modelscited by 160
  25. 2015Interpretable classifiers using rules and Bayesian analysis: Building a better stroke prediction modelcited by 140
  26. 2022Natural Language Descriptions of Deep Visual Featurescited by 136
  27. 2021Adapting Language Models for Zero-shot Learning by Meta-tuning on Dataset and Prompt Collectionscited by 128
  28. 2017A Continuously Growing Dataset of Sentential Paraphrasescited by 72
  29. 2019Analyzing Polarization in Social Media: Method and Application to Tweets on 21 Mass Shootingscited by 67
  30. 2021FLEX: Unifying Evaluation for Few-Shot NLPcited by 67
  31. 2021Prompt Waywardness: The Curious Case of Discretized Interpretation of Continuous Promptscited by 64
  32. 2020Adversarial Semantic Collisionscited by 42
  33. 2014Acoustic Scene Classificationcited by 26
  34. 2019SemEval-2019 Task 8: Fact Checking in Community Question Answering Forumscited by 18
  35. 2019TMLab SRPOL at SemEval-2019 Task 8: Fact Checking in Community Question Answering Forumscited by 2

Built on by

Cited by year (2026 to 7 Oct)

Citing papers per year: 2022: 5; 2023: 10; 2024: 16; 2025: 12; 2026: 11.

54 papers.

Sort:

By area:

  1. 2024Can LLMs Generate Novel Research Ideas? A Large-Scale Human Study with 100+ NLP Researcherscited by 348
  2. 2022Instruction Induction: From Few Examples to Natural Language Task Descriptionscited by 152
  3. 2024Rethinking Interpretability in the Era of Large Language Modelscited by 110
  4. 2023Phenomenal Yet Puzzling: Testing Inductive Reasoning Capabilities of Language Models with Hypothesis Refinementcited by 89
  5. 2025Organize the Web: Constructing Domains Enhances Pre-Training Data Curationcited by 73
  6. 2023Goal Driven Discovery of Distributional Differences via Language Descriptionscited by 61
  7. 2023Explaining black box text modules in natural language with language modelscited by 61
  8. 2023Goal-Driven Explainable Clustering via Language Descriptionscited by 58
  9. 2023Describing Differences in Image Sets with Natural Languagecited by 51
  10. 2025Sleep-time Compute: Beyond Inference Scaling at Test-timecited by 41
  11. 2024Self-Boosting Large Language Models with Synthetic Preference Datacited by 40
  12. 2022Nonparametric Masked Language Modelingcited by 39
38 more
  1. 2024Inverse Constitutional AI: Compressing Preferences into Principlescited by 39
  2. 2023Mass-Producing Failures of Multimodal Systems with Language Modelscited by 36
  3. 2025Search Arena: Analyzing Search-Augmented LLMscited by 30
  4. 2024LLM Comparator: Visual Analytics for Side-by-Side Evaluation of Large Language Modelscited by 29
  5. 2025EvalTree: Profiling Language Model Weaknesses via Hierarchical Capability Treescited by 26
  6. 2024Explaining Datasets in Words: Statistical Models with Natural Language Parameterscited by 21
  7. 2023Effective Human-AI Teams via Learned Natural Language Rules and Onboardingcited by 20
  8. 2023Tree Prompting: Efficient Task Adaptation without Fine-Tuningcited by 17
  9. 2024Concept Bottleneck Models Without Predefined Conceptscited by 17
  10. 2024Preference Optimization with Multi-Sample Comparisonscited by 16
  11. 2024Bayesian Concept Bottleneck Models with LLM Priorscited by 16
  12. 2022Non-Programmers Can Label Programs Indirectly via Active Examples: A Case Study with Text-to-SQLcited by 12
  13. 2024Crafting Interpretable Embeddings by Asking LLMs Questionscited by 11
  14. 2025Learning to Interpret Weight Differences in Language Modelscited by 9
  15. 2025Never Start from Scratch: Expediting On-Device LLM Personalization via Explainable Model Selectioncited by 7
  16. 2025Causal Inference on Outcomes Learned from Textcited by 6
  17. 2022Unsupervised Explanation Generation via Correct Instantiationscited by 5
  18. 2024Generative causal testing to bridge data-driven models and scientific theories in language neurosciencecited by 5
  19. 2023Generating Language Corrections for Teaching Physical Control Taskscited by 4
  20. 2023Survey on AI Ethics: A Socio-technical Perspectivecited by 4
  21. 2024Interpretable Next-token Prediction via the Generalized Induction Headcited by 2
  22. 2025Dataset Featurization: Uncovering Natural Language Features through Unsupervised Data Reconstructioncited by 2
  23. 2024Using Language Models to Disambiguate Lexical Choices in Translationcited by 1
  24. 2025Tell, Don't Show: Leveraging Language Models' Abstractive Retellings to Model Literary Themescited by 1
  25. 2025Deciphering Personalization: Towards Fine-Grained Explainability in Natural Language for Personalized Image Generation Modelscited by 1
  26. 2025Teaching People LLM's Errors and Getting it Rightcited by 1
  27. 2026CyberCertBench: Evaluating LLMs in Cybersecurity Certification Knowledgecited by 1
  28. 2022Explaining Patterns in Data with Language Models via Interpretable Autopromptingcited by 0
  29. 2024Does Geo-co-location Matter? A Case Study of Public Health Conversations during COVID-19cited by 0
  30. 2024Discovering influential text using convolutional neural networkscited by 0
  31. 2024On the Role of Model Prior in Real-World Inductive Reasoningcited by 0
  32. 2025VERBA: Verbalizing Model Differences Using Large Language Modelscited by 0
  33. 2026Discovering Implicit Large Language Model Alignment Objectivescited by 0
  34. 2026"What Are You Really Trying to Do?": Co-Creating Life Goals from Everyday Computer Usecited by 0
  35. 2026LatentDiff: Scaling Semantic Dataset Comparison to Millions of Imagescited by 0
  36. 2026Automatically Finding and Validating Unexpected Side-Effects of Interventions on Language Modelscited by 0
  37. 2026When Hard Negatives Hurt: Bridging the Generative-Discriminative Gap in Hard Negative Synthesis for Retrievalcited by 0
  38. 2026Conditional Hypothesis Generation for LLM-Based Text Analysis with Researcher-Specified Covariatescited by 0

50 most cited of 54 All 54 (JSON)