mlsanityData to 7 Oct 2026

Transformer Language Models Handle Word Frequency in Prediction Head

arXiv 2305.18294 ↗First posted 29 May 2023cs.CLcited by 8

Not on the map. Why?

Abstract (arXiv)

Prediction head is a crucial component of Transformer language models. Despite its direct impact on prediction, this component has often been overlooked in analyzing Transformers. In this study, we investigate the inner workings of the prediction head, specifically focusing on bias parameters. Our experiments with BERT and GPT-2 models reveal that the biases in their word prediction heads play a significant role in the models' ability to reflect word frequency in a corpus, aligning with the logit adjustment method commonly used in long-tailed learning. We also quantify the effect of controlling the biases in practical auto-regressive text generation scenarios; under a particular setting, more diverse text can be generated without compromising text quality.

Key facts. Cited by 8 papers in the map (data through 7 Oct 2026). Builds on 27 of its 38 references that are in the map. First posted on arXiv 29 May 2023.

Builds on

27 of 38 references.

References · 27

  1. 2018BERT: Pre-training of Deep Bidirectional Transformers for Language Understandingcited by 40,204
  2. 2019Exploring the Limits of Transfer Learning with a Unified Text-to-Text Transformercited by 14,491
  3. 2019RoBERTa: A Robustly Optimized BERT Pretraining Approachcited by 13,266
  4. 2022OPT: Open Pre-trained Transformer Language Modelscited by 3,464
  5. 2016Gaussian Error Linear Units (GELUs)cited by 3,431
  6. 2019The Curious Case of Neural Text Degenerationcited by 2,881
  7. 2020Transformer Feed-Forward Layers Are Key-Value Memoriescited by 1,307
  8. 2018Hierarchical Neural Story Generationcited by 1,182
  9. 2015A Diversity-Promoting Objective Function for Neural Conversation Modelscited by 1,068
  10. 2015Aligning Books and Movies: Towards Story-like Visual Explanations by Watching Movies and Reading Bookscited by 1,041
  11. 2019What Does BERT Look At? An Analysis of BERT's Attentioncited by 994
  12. 2020A Primer in BERTology: What we know about how BERT workscited by 807
15 more
  1. 2019How Contextual are Contextualized Word Representations? Comparing the Geometry of BERT, ELMo, and GPT-2 Embeddingscited by 770
  2. 2020Long-tail learning via logit adjustmentcited by 605
  3. 2021Knowledge Neurons in Pretrained Transformerscited by 590
  4. 2021Datasets: A Community Library for Natural Language Processingcited by 526
  5. 2021MAUVE: Measuring the Gap Between Neural Text and Human Text using Divergence Frontierscited by 466
  6. 2017All-but-the-Top: Simple and Effective Postprocessing for Word Representationscited by 205
  7. 2019BPE-Dropout: Simple and Effective Subword Regularizationcited by 189
  8. 2019The Bottom-up Evolution of Representations in the Transformer: A Study with Machine Translation and Language Modeling Objectivescited by 139
  9. 2022Locally Typical Samplingcited by 108
  10. 2021Incorporating Residual and Normalization Layers into Analysis of Masked Language Modelscited by 50
  11. 2018FRAGE: Frequency-Agnostic Word Representationcited by 47
  12. 2021SHAPE: Shifted Absolute Position Embedding for Transformerscited by 37
  13. 2016Reviving Threshold-Moving: a Simple Plug-in Bagging Ensemble for Binary and Multiclass Imbalanced Datacited by 23
  14. 2021Learning to Remove: Towards Isotropic Pre-trained BERT Embeddingcited by 23
  15. 2019From Balustrades to Pierre Vinken: Looking for Syntax in Transformer Self-Attentionscited by 22

Built on by

Cited by year (2026 to 7 Oct)

Citing papers per year: 2024: 5; 2025: 1; 2026: 2.

8 papers.

  1. 2024Do Large Language Models Latently Perform Multi-Hop Reasoning?cited by 165
  2. 2024Interpreting Key Mechanisms of Factual Recall in Transformer-Based Language Modelscited by 30
  3. 2024Understanding Token Probability Encoding in Output Embeddingscited by 6
  4. 2024NoisyICL: A Little Noise in Model Parameters Calibrates In-context Learningcited by 4
  5. 2024Zipfian Whiteningcited by 3
  6. 2025Next-Token Prediction Should be Ambiguity-Sensitive: A Meta-Learning Perspectivecited by 0
  7. 2026Adam's Law: Textual Frequency Law on Large Language Modelscited by 0
  8. 2026The Geometry of Ignorance: LLMs Know When to Temper Bayesian Priorscited by 0