mlsanityData to 14 Oct 2026

Understanding Undesirable Word Embedding Associations

arXiv 1908.06361 ↗First posted 18 Aug 2019cs.CLcited by 59

Not on the map. Why?

Abstract (arXiv)

Word embeddings are often criticized for capturing undesirable word associations such as gender stereotypes. However, methods for measuring and removing such biases remain poorly understood. We show that for any embedding model that implicitly does matrix factorization, debiasing vectors post hoc using subspace projection (Bolukbasi et al., 2016) is, under certain conditions, equivalent to training on an unbiased corpus. We also prove that WEAT, the most common association test for word embeddings, systematically overestimates bias. Given that the subspace projection method is provably effective, we use it to derive a new measure of association called the textitrelational inner product association (RIPA). Experiments with RIPA reveal that, on average, skipgram with negative sampling (SGNS) does not make most words any more gendered than they are in the training corpus. However, for gender-stereotyped words, SGNS actually amplifies the gender association in the corpus.

Key facts. Cited by 59 papers in the map (data through 14 Oct 2026). First posted on arXiv 18 Aug 2019.

Builds on

No reference list yet.

Built on by

Cited by year (2026 to 14 Oct)

Citing papers per year: 2021: 14; 2022: 22; 2023: 11; 2024: 8; 2025: 3; 2026: 1.

59 papers.

Sort:

By area:

  1. 2021RedditBias: A Real-World Resource for Bias Evaluation and Debiasing of Conversational Language Modelscited by 149
  2. 2021A Survey on Gender Bias in Natural Language Processingcited by 89
  3. 2022Theories of "Gender" in NLP Bias Researchcited by 62
  4. 2021Unmasking the Mask -- Evaluating Social Biases in Masked Language Modelscited by 60
  5. 2022Problems with Cosine as a Measure of Embedding Similarity for High Frequency Wordscited by 49
  6. 2022Probing Classifiers are Unreliable for Concept Removal and Detectioncited by 41
  7. 2021Low Frequency Names Exhibit Bias and Overfitting in Contextualizing Language Modelscited by 39
  8. 2022Gold Doesn't Always Glitter: Spectral Removal of Linear and Nonlinear Guarded Attribute Informationcited by 34
  9. 2022MABEL: Attenuating Gender Bias using Textual Entailment Datacited by 31
  10. 2022Choose Your Lenses: Flaws in Gender Bias Evaluationcited by 29
  11. 2023Sociodemographic Bias in Language Models: A Survey and Forward Pathcited by 28
  12. 2021Dictionary-based Debiasing of Pre-trained Word Embeddingscited by 24
38 more
  1. 2021Frequency-based Distortions in Contextualized Word Embeddingscited by 20
  2. 2024Fairness and Bias in Multimodal AI: A Surveycited by 20
  3. 2021Evaluating Gender Bias in Hindi-English Machine Translationcited by 19
  4. 2021Idiosyncratic but not Arbitrary: Learning Idiolects in Online Registers Reveals Distinctive yet Consistent Individual Stylescited by 18
  5. 2022Sense Embeddings are also Biased--Evaluating Social Biases in Static and Contextualised Sense Embeddingscited by 18
  6. 2022Better Hit the Nail on the Head than Beat around the Bush: Removing Protected Attributes with a Single Projectioncited by 17
  7. 2022The Birth of Bias: A case study on the evolution of gender bias in an English language modelcited by 16
  8. 2023Fairness in Language Models Beyond English: Gaps and Challengescited by 16
  9. 2021Assessing the Reliability of Word Embedding Gender Bias Measurescited by 14
  10. 2021Diachronic Analysis of German Parliamentary Proceedings: Ideological Shifts through the Lens of Political Biasescited by 13
  11. 2022How Gender Debiasing Affects Internal Model Representations, and Why It Matterscited by 12
  12. 2023Learning to Generate Equitable Text in Dialogue from Biased Training Datacited by 12
  13. 2021Residual2Vec: Debiasing graph embedding with random graphscited by 11
  14. 2023Comparing Intrinsic Gender Bias Evaluation Measures without using Human Annotated Examplescited by 8
  15. 2021Evaluating Metrics for Bias in Word Embeddingscited by 7
  16. 2022Richer Countries and Richer Representationscited by 7
  17. 2022Trustworthy Social Bias Measurementcited by 7
  18. 2022Mitigating Gender Stereotypes in Hindi and Marathicited by 6
  19. 2022Mind Your Bias: A Critical Review of Bias Detection Methods for Contextual Language Modelscited by 6
  20. 2023Removing Spurious Concepts from Neural Network Representations via Joint Subspace Estimationcited by 6
  21. 2021Theoretical foundations and limits of word embeddings: what types of meaning can they capture?cited by 5
  22. 2022Measuring Gender Bias in Word Embeddings of Gendered Languages Requires Disentangling Grammatical Gender Signalscited by 5
  23. 2022No Word Embedding Model Is Perfect: Evaluating the Representation Accuracy for Social Bias in the Mediacited by 5
  24. 2022Bridging Fairness and Environmental Sustainability in Natural Language Processingcited by 5
  25. 2023Solving Cosine Similarity Underestimation between High Frequency Words by L2 Norm Discountingcited by 5
  26. 2024Statistical Uncertainty in Word Embeddings: GloVe-Vcited by 5
  27. 2021On the Interpretability and Significance of Bias Metrics in Texts: a PMI-based Approachcited by 4
  28. 2022Large scale analysis of gender bias and sexism in song lyricscited by 4
  29. 2022Undesirable Biases in NLP: Addressing Challenges of Measurementcited by 4
  30. 2023A Predictive Factor Analysis of Social Biases and Task-Performance in Pretrained Masked Language Modelscited by 4
  31. 2024Evaluating Short-Term Temporal Fluctuations of Social Biases in Social Media Data and Masked Language Modelscited by 3
  32. 2024ML-EAT: A Multilevel Embedding Association Test for Interpretable and Transparent Social Sciencecited by 3
  33. 2023Investigating the Effects of Fairness Interventions Using Pointwise Representational Similaritycited by 2
  34. 2024Measuring Social Biases in Masked Language Models by Proxy of Prediction Qualitycited by 2
  35. 2025From Charts to Fair Narratives: Uncovering and Mitigating Geo-Economic Biases in Chart-to-Textcited by 2
  36. 2025Are you sure? Measuring models bias in content moderation through uncertaintycited by 2
  37. 2022Regional Negative Bias in Word Embeddings Predicts Racial Animus--but only via Name Frequencycited by 1
  38. 2023Measuring Dimensions of Self-Presentation in Twitter Bios and their Links to Misinformation Sharingcited by 1

50 most cited of 59 All 59 (JSON)