mlsanityData to 7 Oct 2026

Unveiling the Achilles' Heel of NLG Evaluators: A Unified Adversarial Framework Driven by Large Language Models

arXiv 2405.14646 ↗First posted 23 May 2024cs.CLcited by 4

Not on the map. Why?

Abstract (arXiv)

The automatic evaluation of natural language generation (NLG) systems presents a long-lasting challenge. Recent studies have highlighted various neural metrics that align well with human evaluations. Yet, the robustness of these evaluators against adversarial perturbations remains largely under-explored due to the unique challenges in obtaining adversarial data for different NLG evaluation tasks. To address the problem, we introduce AdvEval, a novel black-box adversarial framework against NLG evaluators. AdvEval is specially tailored to generate data that yield strong disagreements between human and victim evaluators. Specifically, inspired by the recent success of large language models (LLMs) in text generation and evaluation, we adopt strong LLMs as both the data generator and gold evaluator. Adversarial data are automatically optimized with feedback from the gold and victim evaluator. We conduct experiments on 12 victim evaluators and 11 NLG datasets, spanning tasks including dialogue, summarization, and question evaluation. The results show that AdvEval can lead to significant performance degradation of various victim metrics, thereby validating its efficacy.

Key facts. Cited by 4 papers in the map (data through 7 Oct 2026). Builds on 56 of its 80 references that are in the map. First posted on arXiv 23 May 2024.

Builds on

56 of 80 references.

References · 56

  1. 2022Training language models to follow instructions with human feedbackcited by 15,712
  2. 2023Judging LLM-as-a-Judge with MT-Bench and Chatbot Arenacited by 8,393
  3. 2019BERTScore: Evaluating Text Generation with BERTcited by 5,462
  4. 2016SQuAD: 100,000+ Questions for Machine Comprehension of Textcited by 4,115
  5. 2018HotpotQA: A Dataset for Diverse, Explainable Multi-hop Question Answeringcited by 3,686
  6. 2023Self-Refine: Iterative Refinement with Self-Feedbackcited by 3,317
  7. 2023G-Eval: NLG Evaluation using GPT-4 with Better Human Alignmentcited by 2,205
  8. 2018A Call for Clarity in Reporting BLEU Scorescited by 1,790
  9. 2018Don't Give Me the Details, Just the Summary! Topic-Aware Convolutional Neural Networks for Extreme Summarizationcited by 1,308
  10. 2017Get To The Point: Summarization with Pointer-Generator Networkscited by 1,281
  11. 2020BLEURT: Learning Robust Metrics for Text Generationcited by 1,196
  12. 2023Teaching Large Language Models to Self-Debugcited by 963
38 more
  1. 2023ChatGPT Outperforms Crowd-Workers for Text-Annotation Taskscited by 849
  2. 2018Personalizing Dialogue Agents: I have a dog, do you have pets too?cited by 820
  3. 2023AlpacaFarm: A Simulation Framework for Methods that Learn from Human Feedbackcited by 738
  4. 2023Can Large Language Models Be an Alternative to Human Evaluations?cited by 729
  5. 2021BARTScore: Evaluating Generated Text as Text Generationcited by 719
  6. 2019Universal Adversarial Triggers for Attacking and Analyzing NLPcited by 709
  7. 2023Baichuan 2: Open Large-scale Language Modelscited by 664
  8. 2019Is BERT Really Robust? A Strong Baseline for Natural Language Attack on Text Classification and Entailmentcited by 663
  9. 2017DailyDialog: A Manually Labelled Multi-turn Dialogue Datasetcited by 649
  10. 2020BERT-ATTACK: Adversarial Attack Against BERT Using BERTcited by 488
  11. 2019MoverScore: Text Generation Evaluating with Contextualized Embeddings and Earth Mover Distancecited by 415
  12. 2018TextBugger: Generating Adversarial Text Against Real-world Applicationscited by 376
  13. 2016How NOT To Evaluate Your Dialogue System: An Empirical Study of Unsupervised Evaluation Metrics for Dialogue Response Generationcited by 356
  14. 2016NewsQA: A Machine Comprehension Datasetcited by 330
  15. 2020BAE: BERT-based Adversarial Examples for Text Classificationcited by 316
  16. 2022ZeroGen: Efficient Zero-shot Learning via Dataset Generationcited by 209
  17. 2022Generating Training Data with Language Models: Towards Zero-Shot Language Understandingcited by 199
  18. 2020USR: An Unsupervised and Reference Free Evaluation Metric for Dialog Generationcited by 186
  19. 2021DialogSum: A Real-Life Scenario Dialogue Summarization Datasetcited by 181
  20. 2023AnnoLLM: Making Large Language Models to Be Better Crowdsourced Annotatorscited by 171
  21. 2020A Survey of Evaluation Metrics Used for NLG Systemscited by 164
  22. 2019DREAM: A Challenge Dataset and Models for Dialogue-Based Reading Comprehensioncited by 156
  23. 2020Unsupervised Evaluation of Interactive Dialog with DialoGPTcited by 139
  24. 2020Contextualized Perturbation for Textual Adversarial Attackcited by 136
  25. 2020MuTual: A Dataset for Multi-Turn Dialogue Reasoningcited by 114
  26. 2021Measure and Improve Robustness in NLP Models: A Surveycited by 107
  27. 2016Crafting Adversarial Input Sequences for Recurrent Neural Networkscited by 102
  28. 2018Seq2Sick: Evaluating the Robustness of Sequence-to-Sequence Models with Adversarial Examplescited by 93
  29. 2021A Comprehensive Assessment of Dialog Evaluation Metricscited by 90
  30. 2017RUBER: An Unsupervised Method for Automatic Evaluation of Open-Domain Dialog Systemscited by 88
  31. 2018On Adversarial Examples for Character-Level Neural Machine Translationcited by 85
  32. 2020GRADE: Automatic Graph-Enhanced Coherence Metric for Evaluating Open-Domain Dialogue Systemscited by 76
  33. 2023TIGERScore: Towards Building Explainable Metric for All Text Generation Taskscited by 69
  34. 2022ProGen: Progressive Zero-shot Dataset Generation via In-context Feedbackcited by 67
  35. 2021DynaEval: Unifying Turn and Dialogue Level Evaluationcited by 59
  36. 2021Perturbation CheckLists for Evaluating NLG Evaluation Metricscited by 49
  37. 2020Designing Precise and Robust Dialogue Response Evaluatorscited by 37
  38. 2021Using Adversarial Attacks to Reveal the Statistical Bias in Machine Reading Comprehension Modelscited by 29

50 most cited of 56 All 56 (JSON)

Built on by

Cited by year (2026 to 7 Oct)

Citing papers per year: 2024: 3; 2025: 0; 2026: 1.

4 papers.

  1. 2024VoiceBench: Benchmarking LLM-Based Voice Assistantscited by 192
  2. 2024Cheating Automatic LLM Benchmarks: Null Models Achieve High Win Ratescited by 42
  3. 2024An LLM Feature-based Framework for Dialogue Constructiveness Assessmentcited by 9
  4. 2026On the Adversarial Robustness of Multimodal LLM Judgescited by 3