mlsanityData to 7 Oct 2026

Fixing Model Bugs with Natural Language Patches

arXiv 2211.03318 ↗First posted 7 Nov 2022cs.CLcited by 32

Not on the map. Why?

Abstract (arXiv)

Current approaches for fixing systematic problems in NLP models (e.g. regex patches, finetuning on more data) are either brittle, or labor-intensive and liable to shortcuts. In contrast, humans often provide corrections to each other through natural language. Taking inspiration from this, we explore natural language patches -- declarative statements that allow developers to provide corrective feedback at the right level of abstraction, either overriding the model (``if a review gives 2 stars, the sentiment is negative'') or providing additional information the model may lack (``if something is described as the bomb, then it is good''). We model the task of determining if a patch applies separately from the task of integrating patch information, and show that with a small amount of synthetic data, we can teach models to effectively use real patches on real data -- 1 to 7 patches improve accuracy by ~1-4 accuracy points on different slices of a sentiment analysis dataset, and F1 by 7 points on a relation extraction dataset. Finally, we show that finetuning on as many as 100 labeled examples may be needed to match the performance of a small set of language patches.

Key facts. Cited by 32 papers in the map (data through 7 Oct 2026). Builds on 20 of its 32 references that are in the map. First posted on arXiv 7 Nov 2022.

Builds on

20 of 32 references.

References · 20

  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. 2021Evaluating Large Language Models Trained on Codecited by 8,053
  4. 2017Billion-scale similarity search with GPUscited by 2,477
  5. 2015Character-level Convolutional Networks for Text Classificationcited by 2,293
  6. 2020Making Pre-trained Language Models Better Few-shot Learnerscited by 1,255
  7. 2020Beyond Accuracy: Behavioral Testing of NLP models with CheckListcited by 979
  8. 2021Editing Factual Knowledge in Language Modelscited by 548
  9. 2021Fast Model Editing at Scalecited by 491
  10. 2018e-SNLI: Natural Language Inference with Natural Language Explanationscited by 468
  11. 2019Using Natural Language for Reward Shaping in Reinforcement Learningcited by 114
  12. 2019FewRel 2.0: Towards More Challenging Few-Shot Relation Classificationcited by 94
8 more
  1. 2017Learning with Latent Languagecited by 75
  2. 2018Training Classifiers with Natural Language Explanationscited by 58
  3. 2020Meta-Learning Requires Meta-Augmentationcited by 57
  4. 2022Improving Intrinsic Exploration with Language Abstractionscited by 49
  5. 2014Learning to Win by Reading Manuals in a Monte-Carlo Frameworkcited by 39
  6. 2018Guiding Policies with Language via Meta-Learningcited by 34
  7. 2020ExpBERT: Representation Engineering with Natural Language Explanationscited by 28
  8. 2019Shaping Visual Representations with Language for Few-shot Classificationcited by 27

Built on by

Cited by year (2026 to 7 Oct)

Citing papers per year: 2022: 1; 2023: 14; 2024: 14; 2025: 2; 2026: 1.

32 papers.

Sort:

By area:

  1. 2023Sparks of Artificial General Intelligence: Early experiments with GPT-4cited by 2,579
  2. 2022Editing Models with Task Arithmeticcited by 1,076
  3. 2023Siren's Song in the AI Ocean: A Survey on Hallucination in Large Language Modelscited by 695
  4. 2023Editing Large Language Models: Problems, Methods, and Opportunitiescited by 367
  5. 2023Automatically Correcting Large Language Models: Surveying the landscape of diverse self-correction strategiescited by 236
  6. 2023Knowledge Editing for Large Language Models: A Surveycited by 189
  7. 2024A Comprehensive Study of Knowledge Editing for Large Language Modelscited by 157
  8. 2024Safe LoRA: the Silver Lining of Reducing Safety Risks when Fine-tuning Large Language Modelscited by 122
  9. 2023RL4F: Generating Natural Language Feedback with Reinforcement Learning for Repairing Model Outputscited by 104
  10. 2024BadEdit: Backdooring large language models by model editingcited by 84
  11. 2023Parameter Efficient Multi-task Model Fusion with Partial Linearizationcited by 64
  12. 2024In-Context Editing: Learning Knowledge from Self-Induced Distributionscited by 33
20 more
  1. 2024RLVF: Learning from Verbal Feedback without Overgeneralizationcited by 18
  2. 2024EVEDIT: Event-based Knowledge Editing with Deductive Editing Boundariescited by 16
  3. 2024Large Language Model Bias Mitigation from the Perspective of Knowledge Editingcited by 16
  4. 2023Forgetting before Learning: Utilizing Parametric Arithmetic for Knowledge Updating in Large Language Modelscited by 15
  5. 2024LEMoE: Advanced Mixture of Experts Adaptor for Lifelong Model Editing of Large Language Modelscited by 14
  6. 2023Robust Weight Signatures: Gaining Robustness as Easy as Patching Weights?cited by 13
  7. 2024Identifying and Mitigating Social Bias Knowledge in Language Modelscited by 13
  8. 2023Let Me Teach You: Pedagogical Foundations of Feedback for Language Modelscited by 9
  9. 2024MEMoE: Enhancing Model Editing with Mixture of Experts Adaptorscited by 8
  10. 2024AI, Meet Human: Learning Paradigms for Hybrid Decision Making Systemscited by 7
  11. 2023Solving NLP Problems through Human-System Collaboration: A Discussion-based Approachcited by 5
  12. 2024Efficiently Quantifying and Mitigating Ripple Effects in Model Editingcited by 5
  13. 2023MaNtLE: Model-agnostic Natural Language Explainercited by 4
  14. 2024Learning Where to Edit Vision Transformerscited by 4
  15. 2024Semantic are Beacons: A Semantic Perspective for Unveiling Parameter-Efficient Fine-Tuning in Knowledge Learningcited by 3
  16. 2026MOMO: Mars Orbital Model Foundation Model for Mars Orbital Applicationscited by 1
  17. 2023Targeted Image Data Augmentation Increases Basic Skills Captioning Robustnesscited by 0
  18. 2023Functionality learning through specification instructionscited by 0
  19. 2025Balancing Knowledge Updates: Toward Unified Modular Editing in LLMscited by 0
  20. 2025Escaping Optimization Stagnation: Taking Steps Beyond Task Arithmetic via Difference Vectorscited by 0