mlsanityData to 13 Oct 2026

Promote, Suppress, Iterate: How Language Models Answer One-to-Many Factual Queries

arXiv 2502.20475 ↗First posted 27 Feb 2025cs.CL cs.AI cs.LGcited by 1

Not on the map. Why?

Abstract (arXiv)

To answer one-to-many factual queries (e.g., listing cities of a country), a language model (LM) must simultaneously recall knowledge and avoid repeating previous answers. How are these two subtasks implemented and integrated internally? Across multiple datasets, models, and prompt templates, we identify a promote-then-suppress mechanism: the model first recalls all answers, and then suppresses previously generated ones. Specifically, LMs use both the subject and previous answer tokens to perform knowledge recall, with attention propagating subject information and MLPs promoting the answers. Then, attention attends to and suppresses previous answer tokens, while MLPs amplify the suppression signal. Our mechanism is corroborated by extensive experimental evidence: in addition to using early decoding and causal tracing, we analyze how components use different tokens by introducing both Token Lens, which decodes aggregated attention updates from specified tokens, and a knockout method that analyzes changes in MLP outputs after removing attention to specified tokens. Overall, we provide new insights into how LMs' internal components interact with different input tokens to support complex factual recall. Code is available at https://github.com/Lorenayannnnn/how-lms-answer-one-to-many-factual-queries.

Key facts. Cited by 1 paper in the map (data through 13 Oct 2026). Builds on 30 of its 37 references that are in the map. First posted on arXiv 27 Feb 2025.

Builds on

30 of 37 references.

References · 30

  1. 2022Locating and Editing Factual Associations in GPTcited by 2,650
  2. 2019Language Models as Knowledge Bases?cited by 1,931
  3. 2019Root Mean Square Layer Normalizationcited by 1,449
  4. 2020Transformer Feed-Forward Layers Are Key-Value Memoriescited by 1,323
  5. 2023Inference-Time Intervention: Eliciting Truthful Answers from a Language Modelcited by 1,280
  6. 2022Interpretability in the Wild: a Circuit for Indirect Object Identification in GPT-2 smallcited by 1,145
  7. 2019Are Sixteen Heads Really Better than One?cited by 807
  8. 2022Discovering Latent Knowledge in Language Models Without Supervisioncited by 750
  9. 2023Eliciting Latent Predictions from Transformers with the Tuned Lenscited by 600
  10. 2021Knowledge Neurons in Pretrained Transformerscited by 597
  11. 2022Transformer Feed-Forward Layers Build Predictions by Promoting Concepts in the Vocabulary Spacecited by 529
  12. 2023Dissecting Recall of Factual Associations in Auto-Regressive Language Modelscited by 517
18 more
  1. 2023Finding Neurons in a Haystack: Case Studies with Sparse Probingcited by 332
  2. 2024Patchscopes: A Unifying Framework for Inspecting Hidden Representations of Language Modelscited by 229
  3. 2023Localizing Model Behavior with Path Patchingcited by 188
  4. 2021Knowledgeable or Educated Guess? Revisiting Language Models as Knowledge Basescited by 107
  5. 2023Circuit Component Reuse Across Tasks in Transformer Language Modelscited by 105
  6. 2023Language Models Implement Simple Word2Vec-style Vector Arithmeticcited by 98
  7. 2023Neurons in Large Language Models: Dead, N-gram, Positionalcited by 88
  8. 2019Inducing Relational Knowledge from BERTcited by 75
  9. 2023Copy Suppression: Comprehensively Understanding an Attention Headcited by 75
  10. 2024LLM Circuit Analyses Are Consistent Across Training and Scalecited by 61
  11. 2024Confidence Regulation Neurons in Language Modelscited by 57
  12. 2022Discovering Latent Concepts Learned in BERTcited by 56
  13. 2024Summing Up the Facts: Additive Mechanisms Behind Factual Recall in LLMscited by 40
  14. 2024Talking Heads: Understanding Inter-layer Communication in Transformer Language Modelscited by 39
  15. 2021Analyzing Commonsense Emergence in Few-shot Knowledge Modelscited by 34
  16. 2023Circuit Breaking: Removing Model Behaviors with Targeted Ablationcited by 27
  17. 2024When Parts Are Greater Than Sums: Individual LLM Components Can Outperform Full Modelscited by 6
  18. 2024Fact Recall, Heuristics or Pure Guesswork? Precise Interpretations of Language Models for Fact Completioncited by 4

Built on by

Cited by year (2026 to 13 Oct)

Citing papers per year: 2026: 1.

1 paper.

  1. 2026How Language Models Process Negationcited by 0