mlsanityData to 7 Oct 2026

BERTScore is Unfair: On Social Bias in Language Model-Based Metrics for Text Generation

arXiv 2210.07626 ↗First posted 14 Oct 2022cs.CL cs.AIcited by 48

Not on the map. Why?

Abstract (arXiv)

Automatic evaluation metrics are crucial to the development of generative systems. In recent years, pre-trained language model (PLM) based metrics, such as BERTScore, have been commonly adopted in various generation tasks. However, it has been demonstrated that PLMs encode a range of stereotypical societal biases, leading to a concern on the fairness of PLMs as metrics. To that end, this work presents the first systematic study on the social bias in PLM-based metrics. We demonstrate that popular PLM-based metrics exhibit significantly higher social bias than traditional metrics on 6 sensitive attributes, namely race, gender, religion, physical appearance, age, and socioeconomic status. In-depth analysis suggests that choosing paradigms (matching, regression, or generation) of the metric has a greater impact on fairness than choosing PLMs. In addition, we develop debiasing adapters that are injected into PLM layers, mitigating bias in PLM-based metrics while retaining high performance for evaluating text generation.

Key facts. Cited by 48 papers in the map (data through 7 Oct 2026). Builds on 38 of its 52 references that are in the map. First posted on arXiv 14 Oct 2022.

Builds on

38 of 52 references.

References · 38

  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. 2015Distilling the Knowledge in a Neural Networkcited by 10,076
  4. 2019BART: Denoising Sequence-to-Sequence Pre-training for Natural Language Generation, Translation, and Comprehensioncited by 5,874
  5. 2019BERTScore: Evaluating Text Generation with BERTcited by 5,462
  6. 2019DistilBERT, a distilled version of BERT: smaller, faster, cheaper and lightercited by 4,243
  7. 2019Parameter-Efficient Transfer Learning for NLPcited by 4,195
  8. 2019XLNet: Generalized Autoregressive Pretraining for Language Understandingcited by 2,801
  9. 2019ALBERT: A Lite BERT for Self-supervised Learning of Language Representationscited by 2,772
  10. 2017A Broad-Coverage Challenge Corpus for Sentence Understanding through Inferencecited by 2,665
  11. 2020BLEURT: Learning Robust Metrics for Text Generationcited by 1,196
  12. 2020COMET: A Neural Framework for MT Evaluationcited by 1,024
26 more
  1. 2017SemEval-2017 Task 1: Semantic Textual Similarity - Multilingual and Cross-lingual Focused Evaluationcited by 998
  2. 2020StereoSet: Measuring stereotypical bias in pretrained language modelscited by 911
  3. 2020AdapterFusion: Non-Destructive Task Composition for Transfer Learningcited by 780
  4. 2021BARTScore: Evaluating Generated Text as Text Generationcited by 719
  5. 2020CrowS-Pairs: A Challenge Dataset for Measuring Social Biases in Masked Language Modelscited by 703
  6. 2018Gender Bias in Coreference Resolution: Evaluation and Debiasing Methodscited by 657
  7. 2020Pre-trained Models for Natural Language Processing: A Surveycited by 504
  8. 2020AdapterHub: A Framework for Adapting Transformerscited by 441
  9. 2019MoverScore: Text Generation Evaluating with Contextualized Embeddings and Earth Mover Distancecited by 415
  10. 2019On Measuring Social Biases in Sentence Encoderscited by 397
  11. 2019Measuring Bias in Contextualized Word Representationscited by 275
  12. 2020Measuring and Reducing Gendered Correlations in Pre-trained Modelscited by 222
  13. 2019Identifying and Reducing Gender Bias in Word-Level Language Modelscited by 214
  14. 2019Gender Bias in Contextualized Word Embeddingscited by 203
  15. 2021RedditBias: A Real-World Resource for Bias Evaluation and Debiasing of Conversational Language Modelscited by 149
  16. 2020Re-evaluating Evaluation in Text Summarizationcited by 144
  17. 2019Assessing Social and Intersectional Biases in Contextualized Word Representationscited by 134
  18. 2020Automatic Machine Translation Evaluation in Many Languages via Zero-Shot Paraphrasingcited by 114
  19. 2019On Measuring and Mitigating Biased Inferences of Word Embeddingscited by 112
  20. 2020Rethinking embedding coupling in pre-trained language modelscited by 103
  21. 2021Sustainable Modular Debiasing of Language Modelscited by 98
  22. 2019Evaluating the Underlying Gender Bias in Contextualized Word Embeddingscited by 94
  23. 2021Learning Compact Metrics for MTcited by 93
  24. 2021Quantifying Social Biases in NLP: A Generalization and Empirical Comparison of Extrinsic Fairness Metricscited by 83
  25. 2021Unmasking the Mask -- Evaluating Social Biases in Masked Language Modelscited by 60
  26. 2021FrugalScore: Learning Cheaper, Lighter and Faster Evaluation Metricsfor Automatic Text Generationcited by 19

Built on by

Cited by year (2026 to 7 Oct)

Citing papers per year: 2023: 16; 2024: 14; 2025: 14; 2026: 4.

48 papers.

Sort:

By area:

  1. 2024From Generation to Judgment: Opportunities and Challenges of LLM-as-a-judgecited by 480
  2. 2024Measuring Political Bias in Large Language Models: What Is Said and How It Is Saidcited by 89
  3. 2024A Survey on Multilingual Large Language Models: Corpora, Alignment, and Biascited by 82
  4. 2024LLM-based NLG Evaluation: Current Status and Challengescited by 71
  5. 2023Corex: Pushing the Boundaries of Complex Reasoning through Multi-Model Collaborationcited by 70
  6. 2024Fairness in Large Language Models: A Taxonomic Surveycited by 57
  7. 2023DERA: Enhancing Large Language Model Completions with Dialog-Enabled Resolving Agentscited by 52
  8. 2025Neither Valid nor Reliable? Investigating the Use of LLMs as Judgescited by 37
  9. 2023Investigating Subtler Biases in LLMs: Ageism, Beauty, Institutional, and Nationality Bias in Generative Modelscited by 32
  10. 2023Elastic Weight Removal for Faithful and Abstractive Dialogue Generationcited by 29
  11. 2024Reference-Guided Verdict: LLMs-as-Judges in Automatic Evaluation of Free-Form QAcited by 25
  12. 2023Gender Biases in Automatic Evaluation Metrics for Image Captioningcited by 24
36 more
  1. 2023BLEURT Has Universal Translations: An Analysis of Automatic Metrics by Minimum Risk Trainingcited by 24
  2. 2024Can LLMs Recognize Toxicity? A Structured Investigation Framework and Toxicity Metriccited by 23
  3. 2024Fine-Tuned Machine Translation Metrics Struggle in Unseen Domainscited by 21
  4. 2025Llama-3.1-FoundationAI-SecurityLLM-8B-Instruct Technical Reportcited by 21
  5. 2023NLG Evaluation Metrics Beyond Correlation Analysis: An Empirical Metric Preference Checklistcited by 16
  6. 2023Towards Explainable Evaluation Metrics for Machine Translationcited by 15
  7. 2023An Appraisal-Based Chain-Of-Emotion Architecture for Affective Language Model Game Agentscited by 15
  8. 2024Position: Towards Bidirectional Human-AI Alignmentcited by 15
  9. 2024Take Care of Your Prompt Bias! Investigating and Mitigating Prompt Bias in Factual Knowledge Extractioncited by 13
  10. 2024A Comparative Study of Quality Evaluation Methods for Text Summarizationcited by 13
  11. 2023ContraDoc: Understanding Self-Contradictions in Documents with Large Language Modelscited by 12
  12. 2023Overview of Robust and Multilingual Automatic Evaluation Metrics for Open-Domain Dialogue Systems at DSTC 11 Track 4cited by 10
  13. 2023AMRFact: Enhancing Summarization Factuality Evaluation with AMR-Driven Negative Samples Generationcited by 10
  14. 2024Watching the Watchers: Exposing Gender Disparities in Machine Translation Quality Estimationcited by 10
  15. 2025TituLLMs: A Family of Bangla LLMs with Comprehensive Benchmarkingcited by 9
  16. 2025Steering Towards Fairness: Mitigating Political Bias in LLMscited by 7
  17. 2025Framing Political Bias in Multilingual LLMs Across Pakistani Languagescited by 5
  18. 2025Intent-Aware Schema Generation And Refinement For Literature Review Tablescited by 5
  19. 2023Defining a New NLP Playgroundcited by 4
  20. 2025GAMBIT+: A Challenge Set for Evaluating Gender Bias in Machine Translation Quality Estimation Metricscited by 4
  21. 2023That was the last straw, we need more: Are Translation Systems Sensitive to Disambiguating Context?cited by 3
  22. 2024Measuring Retrieval Complexity in Question Answering Systemscited by 3
  23. 2025BatchGEMBA: Token-Efficient Machine Translation Evaluation with Batched Prompting and Prompt Compressioncited by 3
  24. 2026VEAT Quantifies Implicit Associations in Text-to-Video Generator Sora and Reveals Challenges in Bias Mitigationcited by 2
  25. 2024Expert-Guided Extinction of Toxic Tokens for Debiased Generationcited by 1
  26. 2025ContrastScore: Towards Higher Quality, Less Biased, More Efficient Evaluation Metrics with Contrastive Evaluationcited by 1
  27. 2025From the Laboratory to Real-World Application: Evaluating Zero-Shot Scene Interpretation on Edge Devices for Mobile Roboticscited by 1
  28. 2025Music Recommendation with Large Language Models: Challenges, Opportunities, and Evaluationcited by 1
  29. 2026Embedded Conditional Independence Tests for Large Language Model Generated Text with an Application to German Parliament Speechescited by 1
  30. 2023Dior-CVAE: Pre-trained Language Models and Diffusion Priors for Variational Dialog Generationcited by 0
  31. 2023Reference Free Domain Adaptation for Translation of Noisy Questions with Question Specific Rewardscited by 0
  32. 2025The illusion of a perfect metric: Why evaluating AI's words is harder than it lookscited by 0
  33. 2025Asm2SrcEval: Evaluating Large Language Models for Assembly-to-Source Code Translationcited by 0
  34. 2025Comparing Approaches to Automatic Summarization in Less-Resourced Languagescited by 0
  35. 2026LLM-ReSum: A Framework for LLM Reflective Summarization through Self-Evaluationcited by 0
  36. 2026A Comparative Study of Controlled Text Generation Systems Using Level-Playing-Field Evaluation Principlescited by 0