mlsanityData to 14 Oct 2026

Probing BERT's priors with serial reproduction chains

arXiv 2202.12226 ↗First posted 24 Feb 2022cs.CLcited by 11

Not on the map. Why?

Abstract (arXiv)

Sampling is a promising bottom-up method for exposing what generative models have learned about language, but it remains unclear how to generate representative samples from popular masked language models (MLMs) like BERT. The MLM objective yields a dependency network with no guarantee of consistent conditional distributions, posing a problem for naive approaches. Drawing from theories of iterated learning in cognitive science, we explore the use of serial reproduction chains to sample from BERT's priors. In particular, we observe that a unique and consistent estimator of the ground-truth joint distribution is given by a Generative Stochastic Network (GSN) sampler, which randomly selects which token to mask and reconstruct on each step. We show that the lexical and syntactic statistics of sentences from GSN chains closely match the ground-truth corpus distribution and perform better than other methods in a large corpus of naturalness judgments. Our findings establish a firmer theoretical foundation for bottom-up probing and highlight richer deviations from human priors.

Key facts. Cited by 11 papers in the map (data through 14 Oct 2026). Builds on 32 of its 57 references that are in the map. First posted on arXiv 24 Feb 2022.

Builds on

32 of 57 references.

References · 32

  1. 2018BERT: Pre-training of Deep Bidirectional Transformers for Language Understandingcited by 40,399
  2. 2020Language Models are Few-Shot Learnerscited by 32,419
  3. 2019BART: Denoising Sequence-to-Sequence Pre-training for Natural Language Generation, Translation, and Comprehensioncited by 5,893
  4. 2019The Curious Case of Neural Text Degenerationcited by 2,916
  5. 2019BERT Rediscovers the Classical NLP Pipelinecited by 1,059
  6. 2018A Survey Of Methods For Explaining Black Box Modelscited by 973
  7. 2018Neural Network Acceptability Judgmentscited by 958
  8. 2016Semantics derived automatically from language corpora contain human-like biasescited by 901
  9. 2019BLiMP: The Benchmark of Linguistic Minimal Pairs for Englishcited by 564
  10. 2021MAUVE: Measuring the Gap Between Neural Text and Human Text using Divergence Frontierscited by 470
  11. 2019What BERT is not: Lessons from a new suite of psycholinguistic diagnostics for language modelscited by 320
  12. 2021All That's 'Human' Is Not Gold: Evaluating Human Evaluation of Generated Textcited by 287
20 more
  1. 2020Evaluation of Text Generation: A Surveycited by 277
  2. 2019Automatic Detection of Generated Text is Easiest when Humans are Fooledcited by 261
  3. 2017Word Embeddings Quantify 100 Years of Gender and Ethnic Stereotypescited by 253
  4. 2018Colorless green recurrent networks dream hierarchicallycited by 213
  5. 2019BERT has a Mouth, and It Must Speak: BERT as a Markov Random Field Language Modelcited by 185
  6. 2013Generalized Denoising Auto-Encoders as Generative Modelscited by 113
  7. 2019Don't Say That! Making Inconsistent Dialogue Unlikely with Unlikelihood Trainingcited by 110
  8. 2020Syntactic Structure from Deep Learningcited by 99
  9. 2018Language GANs Falling Shortcited by 97
  10. 2021Language Model Evaluation Beyond Perplexitycited by 78
  11. 2021Lower Perplexity is Not Always Human-Likecited by 52
  12. 2021Exposing the Implicit Energy Networks behind Masked Language Models via Metropolis--Hastingscited by 37
  13. 2013Deep Generative Stochastic Networks Trainable by Backpropcited by 36
  14. 2019Calibration, Entropy Rates, and Memory in Language Modelscited by 32
  15. 2020Gibbs Sampling with Peoplecited by 28
  16. 2019Evaluating Computational Language Models with Scaling Properties of Natural Languagecited by 22
  17. 2020Investigating representations of verb bias in neural language modelscited by 22
  18. 2017Do Neural Nets Learn Statistical Laws behind Natural Language?cited by 16
  19. 2019Analyzing machine-learned representations: A natural language case studycited by 5
  20. 2021Evaluating Models of Robust Word Recognition with Serial Reproductioncited by 2

Built on by

Cited by year (2026 to 14 Oct)

Citing papers per year: 2022: 2; 2023: 4; 2024: 3; 2025: 1; 2026: 1.

11 papers.

  1. 2023Amortizing intractable inference in large language modelscited by 97
  2. 2024QUEST: Quality-Aware Metropolis-Hastings Sampling for Machine Translationcited by 16
  3. 2024What Should Embeddings Embed? Autoregressive Models Represent Latent Generating Distributionscited by 10
  4. 2023Deriving Language Models from Masked Language Modelscited by 9
  5. 2022Testing the limits of natural language models for predicting human language judgmentscited by 7
  6. 2023Deep de Finetti: Recovering Topic Distributions from Large Language Modelscited by 7
  7. 2022Inconsistencies in Masked Language Modelscited by 2
  8. 2025Do Language Models Have Bayesian Brains? Distinguishing Stochastic and Deterministic Decision Patterns within Large Language Modelscited by 2
  9. 2023Bidirectional Transformer Reranker for Grammatical Error Correctioncited by 1
  10. 2024Quantifying Human Priors over Social and Navigation Networkscited by 1
  11. 2026Generative AI & Fictionality: How Novels Power Large Language Modelscited by 1