mlsanityData to 14 Oct 2026

On the Blind Spots of Model-Based Evaluation Metrics for Text Generation

arXiv 2212.10020 ↗First posted 20 Dec 2022cs.CLcited by 49

Not on the map. Why?

Abstract (arXiv)

In this work, we explore a useful but often neglected methodology for robustness analysis of text generation evaluation metrics: stress tests with synthetic data. Basically, we design and synthesize a wide range of potential errors and check whether they result in a commensurate drop in the metric scores. We examine a range of recently proposed evaluation metrics based on pretrained language models, for the tasks of open-ended generation, translation, and summarization. Our experiments reveal interesting insensitivities, biases, or even loopholes in existing metrics. For example, we find that BERTScore is confused by truncation errors in summarization, and MAUVE (built on top of GPT-2) is insensitive to errors at the beginning or middle of generations. Further, we investigate the reasons behind these blind spots and suggest practical workarounds for a more reliable evaluation of text generation. We have released our code and data at https://github.com/cloudygoose/blindspot_nlg.

Key facts. Cited by 49 papers in the map (data through 14 Oct 2026). Builds on 69 of its 89 references that are in the map. First posted on arXiv 20 Dec 2022.

Builds on

69 of 89 references.

References · 69

  1. 2018BERT: Pre-training of Deep Bidirectional Transformers for Language Understandingcited by 40,399
  2. 2019Exploring the Limits of Transfer Learning with a Unified Text-to-Text Transformercited by 14,599
  3. 2019RoBERTa: A Robustly Optimized BERT Pretraining Approachcited by 13,372
  4. 2019BART: Denoising Sequence-to-Sequence Pre-training for Natural Language Generation, Translation, and Comprehensioncited by 5,893
  5. 2019BERTScore: Evaluating Text Generation with BERTcited by 5,512
  6. 2021Finetuned Language Models Are Zero-Shot Learnerscited by 3,780
  7. 2022OPT: Open Pre-trained Transformer Language Modelscited by 3,512
  8. 2020Longformer: The Long-Document Transformercited by 3,413
  9. 2016Pointer Sentinel Mixture Modelscited by 3,096
  10. 2019The Curious Case of Neural Text Degenerationcited by 2,916
  11. 2021CLIPScore: A Reference-free Evaluation Metric for Image Captioningcited by 2,424
  12. 2020BLEURT: Learning Robust Metrics for Text Generationcited by 1,203
38 more
  1. 2018Hierarchical Neural Story Generationcited by 1,193
  2. 2015Teaching Machines to Read and Comprehendcited by 1,049
  3. 2020COMET: A Neural Framework for MT Evaluationcited by 1,033
  4. 2020Beyond Accuracy: Behavioral Testing of NLP models with CheckListcited by 1,006
  5. 2019Right for the Wrong Reasons: Diagnosing Syntactic Heuristics in Natural Language Inferencecited by 876
  6. 2021Probing Classifiers: Promises, Shortcomings, and Advancescited by 876
  7. 2021BARTScore: Evaluating Generated Text as Text Generationcited by 722
  8. 2021Cross-Task Generalization via Natural Language Crowdsourcing Instructionscited by 699
  9. 2019How multilingual is Multilingual BERT?cited by 645
  10. 2019Evaluating the Factual Consistency of Abstractive Text Summarizationcited by 584
  11. 2019Neural Text Generation with Unlikelihood Trainingcited by 503
  12. 2020BERT-ATTACK: Adversarial Attack Against BERT Using BERTcited by 491
  13. 2021MAUVE: Measuring the Gap Between Neural Text and Human Text using Divergence Frontierscited by 470
  14. 2021DExperts: Decoding-Time Controlled Text Generation with Experts and Anti-Expertscited by 441
  15. 2019MoverScore: Text Generation Evaluating with Contextualized Embeddings and Earth Mover Distancecited by 418
  16. 2020Asking and Answering Questions to Evaluate the Factual Consistency of Summariescited by 378
  17. 2021FUDGE: Controlled Text Generation With Future Discriminatorscited by 376
  18. 2019What BERT is not: Lessons from a new suite of psycholinguistic diagnostics for language modelscited by 320
  19. 2021Understanding Factuality in Abstractive Summarization with FRANK: A Benchmark for Factuality Metricscited by 287
  20. 2018Analysis Methods in Neural Language Processing: A Surveycited by 277
  21. 2020Evaluation of Text Generation: A Surveycited by 277
  22. 2022Towards a Unified Multi-Dimensional Evaluator for Text Generationcited by 264
  23. 2019Analyzing the Structure of Attention in a Transformer Language Modelcited by 243
  24. 2021The GEM Benchmark: Natural Language Generation, its Evaluation and Metricscited by 239
  25. 2018Stress Test Evaluation for Natural Language Inferencecited by 213
  26. 2020USR: An Unsupervised and Reference Free Evaluation Metric for Dialog Generationcited by 188
  27. 2021XTREME-R: Towards More Challenging and Nuanced Multilingual Evaluationcited by 168
  28. 2020Tangled up in BLEU: Reevaluating the Evaluation of Automatic Machine Translation Evaluation Metricscited by 158
  29. 2018Generating Informative and Diverse Conversational Responses via Adversarial Information Maximizationcited by 116
  30. 2020Automatic Machine Translation Evaluation in Many Languages via Zero-Shot Paraphrasingcited by 115
  31. 2018Sharp Nearby, Fuzzy Far Away: How Neural Language Models Use Contextcited by 113
  32. 2021Robustness Gym: Unifying the NLP Evaluation Landscapecited by 108
  33. 2017Neural Machine Translation with Extended Contextcited by 95
  34. 2018Seq2Sick: Evaluating the Robustness of Sequence-to-Sequence Models with Adversarial Examplescited by 94
  35. 2021Survey of Low-Resource Machine Translationcited by 92
  36. 2020GO FIGURE: A Meta Evaluation of Factuality in Summarizationcited by 76
  37. 2019Perturbation Sensitivity Analysis to Detect Unintended Model Biasescited by 74
  38. 2020Out of Order: How Important Is The Sequential Order of Words in a Sentence in Natural Language Understanding Tasks?cited by 73

50 most cited of 69 All 69 (JSON)

Built on by

Cited by year (2026 to 14 Oct)

Citing papers per year: 2023: 15; 2024: 15; 2025: 10; 2026: 9.

49 papers.

Sort:

By area:

  1. 2023Large Language Models are not Fair Evaluatorscited by 1,085
  2. 2024MT-Bench-101: A Fine-Grained Benchmark for Evaluating Large Language Models in Multi-Turn Dialoguescited by 250
  3. 2023LLMs as Narcissistic Evaluators: When Ego Inflates Evaluation Scorescited by 90
  4. 2024LLM-based NLG Evaluation: Current Status and Challengescited by 73
  5. 2023CritiqueLLM: Towards an Informative Critique Generation Model for Evaluation of Large Language Model Generationcited by 66
  6. 2023Uncertainty in Natural Language Generation: From Theory to Applicationscited by 61
  7. 2023Understanding In-Context Learning via Supportive Pretraining Datacited by 41
  8. 2024Are LLM-based Evaluators Confusing NLG Quality Criteria?cited by 36
  9. 2023BLESS: Benchmarking Large Language Models on Sentence Simplificationcited by 35
  10. 2024Finding Blind Spots in Evaluator LLMs with Interpretable Checklistscited by 32
  11. 2023Do GPTs Produce Less Literal Translations?cited by 30
  12. 2024Stumbling Blocks: Stress Testing the Robustness of Machine-Generated Text Detectors Under Attackscited by 26
37 more
  1. 2024SELF-[IN]CORRECT: LLMs Struggle with Discriminating Self-Generated Responsescited by 26
  2. 2023The Eval4NLP 2023 Shared Task on Prompting Large Language Models as Explainable Metricscited by 23
  3. 2024Automatic Metrics in Natural Language Generation: A Survey of Current Evaluation Practicescited by 20
  4. 2024Text Generation: A Systematic Literature Review of Tasks, Evaluation, and Challengescited by 18
  5. 2025Evaluating LLM Metrics Through Real-World Capabilitiescited by 18
  6. 2024What Makes a Good Story and How Can We Measure It? A Comprehensive Survey of Story Evaluationcited by 17
  7. 2023APPLS: Evaluating Evaluation Metrics for Plain Language Summarizationcited by 15
  8. 2023Towards Explainable Evaluation Metrics for Machine Translationcited by 15
  9. 2023ALLURE: Auditing and Improving LLM-based Evaluation of Text using Iterative In-Context-Learningcited by 15
  10. 2024Diffusion Guided Language Modelingcited by 15
  11. 2024Verifiable by Design: Aligning Language Models to Quote from Pre-Training Datacited by 13
  12. 2023DeltaScore: Fine-Grained Story Evaluation with Perturbationscited by 9
  13. 2024Towards Better Open-Ended Text Generation: A Multicriteria Evaluation Frameworkcited by 9
  14. 2025SynthTextEval: Synthetic Text Data Generation and Evaluation for High-Stakes Domainscited by 7
  15. 2023Is Summary Useful or Not? An Extrinsic Human Evaluation of Text Summaries on Downstream Taskscited by 6
  16. 2025IberBench: LLM Evaluation on Iberian Languagescited by 5
  17. 2024An In-depth Evaluation of Large Language Models in Sentence Simplification with Error-based Human Assessmentcited by 4
  18. 2023Learning Human-Human Interactions in Images from Weak Textual Supervisioncited by 3
  19. 2024Towards Reliable Detection of LLM-Generated Texts: A Comprehensive Evaluation Framework with CUDRTcited by 3
  20. 2024On the Low-Rank Parametrization of Reward Models for Controlled Language Generationcited by 3
  21. 2025Contextual Metric Meta-Evaluation by Measuring Local Metric Accuracycited by 3
  22. 2025CAPO: Confidence Aware Preference Optimization Learning for Multilingual Preferencescited by 3
  23. 2023HANS, are you clever? Clever Hans Effect Analysis of Neural Systemscited by 2
  24. 2025Calibrating MLLM-as-a-judge via Multimodal Bayesian Prompt Ensemblescited by 2
  25. 2026Prototypicality Bias Reveals Blindspots in Multimodal Evaluation Metricscited by 2
  26. 2026CE-RM: A Pointwise Generative Reward Model Optimized via Two-Stage Rollout and Unified Criteriacited by 2
  27. 2025A Dual-Perspective NLG Meta-Evaluation Framework with Automatic Benchmark and Better Interpretabilitycited by 1
  28. 2025Can one size fit all?: Measuring Failure in Multi-Document Summarization Domain Transfercited by 1
  29. 2026MATCHA: Matching Text via Contrastive Semantic Alignmentcited by 1
  30. 2026On The Effectiveness-Fluency Trade-Off In LLM Conditioning: A Systematic Studycited by 1
  31. 2025Metric assessment protocol in the context of answer fluctuation on MCQ taskscited by 0
  32. 2025Direct-Scoring NLG Evaluators Can Use Pairwise Comparisons Toocited by 0
  33. 2026A Comparative Study of Controlled Text Generation Systems Using Level-Playing-Field Evaluation Principlescited by 0
  34. 2026Reducing Hallucinations in Complex Question Answering using Simple Graph-based Retrieval-Augmented Generation (long version)cited by 0
  35. 2026Reference-Based Analysis of Coherence and Diversity in Open-Ended Text Generationcited by 0
  36. 2026Coherence-Aware Distributional Evaluation of Open-Ended Text Generationcited by 0
  37. 2026Likelihood Is Not Authorship: Can We Reliably Measure Human Contribution in AI-Assisted Writing?cited by 0