mlsanityData to 14 Oct 2026

When Choosing Plausible Alternatives, Clever Hans can be Clever

arXiv 1911.00225 ↗First posted 1 Nov 2019cs.CLcited by 40

Not on the map. Why?

Abstract (arXiv)

Pretrained language models, such as BERT and RoBERTa, have shown large improvements in the commonsense reasoning benchmark COPA. However, recent work found that many improvements in benchmarks of natural language understanding are not due to models learning the task, but due to their increasing ability to exploit superficial cues, such as tokens that occur more often in the correct answer than the wrong one. Are BERT's and RoBERTa's good performance on COPA also caused by this? We find superficial cues in COPA, as well as evidence that BERT exploits these cues. To remedy this problem, we introduce Balanced COPA, an extension of COPA that does not suffer from easy-to-exploit single token cues. We analyze BERT's and RoBERTa's performance on original and Balanced COPA, finding that BERT relies on superficial cues when they are present, but still achieves comparable performance once they are made ineffective, suggesting that BERT learns the task to a certain degree when forced to. In contrast, RoBERTa does not appear to rely on superficial cues.

Key facts. Cited by 40 papers in the map (data through 14 Oct 2026). First posted on arXiv 1 Nov 2019.

Builds on

No reference list yet.

Built on by

Cited by year (2026 to 14 Oct)

Citing papers per year: 2021: 7; 2022: 3; 2023: 9; 2024: 6; 2025: 9; 2026: 6.

40 papers.

Sort:

By area:

  1. 2023Clever Hans or Neural Theory of Mind? Stress Testing Social Reasoning in Large Language Modelscited by 163
  2. 2025SuperBPE: Space Travel for Language Modelscited by 63
  3. 2023Mind the instructions: a holistic evaluation of consistency and interactions in prompt-based learningcited by 31
  4. 2024Language Models Represent Beliefs of Self and Otherscited by 31
  5. 2023It Takes Two to Tango: Navigating Conceptualizations of NLP Tasks and Measurements of Performancecited by 26
  6. 2023COLA: Contextualized Commonsense Causal Reasoning from the Causal Inference Perspectivecited by 22
  7. 2024FaithLM: Towards Faithful Explanations for Large Language Modelscited by 19
  8. 2025SEA-HELM: Southeast Asian Holistic Evaluation of Language Modelscited by 19
  9. 2022ROCK: Causal Inference Principles for Reasoning about Commonsense Causalitycited by 16
  10. 2021CausalBERT: Injecting Causal Knowledge Into Pre-trained Models with Minimal Supervisioncited by 12
  11. 2022Are Prompt-based Models Clueless?cited by 11
  12. 2024Are LLMs Socially Adaptive? Contrasting Belief Evolution in Large Language Models and Humanscited by 11
28 more
  1. 2024mCSQA: Multilingual Commonsense Reasoning Dataset with Unified Creation Strategy by Language Models and Humanscited by 10
  2. 2021Improving Commonsense Causal Reasoning by Adversarial Training and Data Augmentationcited by 9
  3. 2024Leaving the barn door open for Clever Hans: Simple features predict LLM benchmark answerscited by 9
  4. 2022COPA-SSE: Semi-structured Explanations for Commonsense Reasoningcited by 8
  5. 2021Knowledge-Augmented Language Models for Cause-Effect Relation Classificationcited by 7
  6. 2025CAPO: Cost-Aware Prompt Optimizationcited by 7
  7. 2021Doing Good or Doing Right? Exploring the Weakness of Commonsense Causal Reasoning Modelscited by 5
  8. 2023Average Is Not Enough: Caveats of Multilingual Evaluationcited by 5
  9. 2025Batayan: A Filipino NLP benchmark for evaluating Large Language Modelscited by 5
  10. 2021Learning to Learn to be Right for the Right Reasonscited by 4
  11. 2025Improving the OOD Performance of Closed-Source LLMs on NLI Through Strategic Data Selectioncited by 4
  12. 2026NITP: Next Implicit Token Prediction for LLM Pre-trainingcited by 4
  13. 2021Generating Hypothetical Events for Abductive Inferencecited by 3
  14. 2025AgentCoMa: A Compositional Benchmark Mixing Commonsense and Mathematical Reasoning in Real-World Scenarioscited by 3
  15. 2023Text-To-KG Alignment: Comparing Current Methods on Classification Taskscited by 2
  16. 2023GLS-CSC: A Simple but Effective Strategy to Mitigate Chinese STM Models' Over-Reliance on Superficial Cluecited by 2
  17. 2023INSPECT: Intrinsic and Systematic Probing Evaluation for Code Transformerscited by 2
  18. 2023Out-of-Distribution Generalization in Text Classification: Past, Present, and Futurecited by 1
  19. 2024Enabling Natural Zero-Shot Prompting on Encoder Models via Statement-Tuningcited by 1
  20. 2025Scaling, Simplification, and Adaptation: Lessons from Pretraining on Machine-Translated Textcited by 1
  21. 2025Vocab Diet: Reshaping the Vocabulary of LLMs via Vector Arithmeticcited by 1
  22. 2026BURMESE-SAN: Burmese NLP Benchmark for Evaluating Large Language Modelscited by 1
  23. 2021Go Beyond Plain Fine-tuning: Improving Pretrained Models for Social Commonsensecited by 0
  24. 2025Culturally Grounded Physical Commonsense Reasoning in Italian and English: A Submission to the MRL 2025 Shared Taskcited by 0
  25. 2026Knowing Bias, Doing Better: Mitigating Social Bias in LLMs via Know-Bias Neuron Enhancementcited by 0
  26. 2026A Study on Question-Answer Dataset for LLM Safety Evaluation with a Focus on Illegal Activitiescited by 0
  27. 2026Debias It Yourself: Teaching LLMs Cognitive Bias Mitigation Interventionscited by 0
  28. 2026More Than Words: Compositional Tokenization for Efficient Language Modelscited by 0