Elamur Research⌕
☰
home / papers / 2023 / Prompting for explanations improves Adversarial NLI. Is this true? \Yes\ it is \true\ because ıt weakens superficial cues\

Prompting for explanations improves Adversarial NLI. Is this true? \Yes\ it is \true\ because ıt weakens superficial cues\

Paper · 2023 · Pride Kavumba, Ana Brassard, Benjamin Heinzerling et al.

Summary

The paper asks whether making a model justify its NLI decisions in free text does more than make it interpretable. Specifically, it tests whether explanations also make the model less dependent on shallow lexical shortcuts. The authors fine-tune generative models (T0 and T5-3B) on a third of the e-SNLI and ANLI training data. They compare two output formats. One produces only a label. The other produces a yes/no answer, then the label, then a justification. Both are evaluated on six benchmarks: e-SNLI, ANLI, SNLI Hard, NLI Diagnostic, HANS and Counter-NLI. Explanation prompting beats label-only prompting on almost all of them and exceeds the reported state of the art on ANLI, Counter-NLI and SNLI-Hard. The gains hold across architectures, sizes and prompt wordings. An association analysis based on PPMI between hypothesis tokens and labels shows that words which strongly signal a label in the label-only setup lose that signal once explanations are required. Crowd workers judged the generated explanations about as good as human ones. The work suggests a cheap way to gain robustness without adversarial training, provided explanation annotations exist.

Key points

  • The output template is "{Yes/No} it is {label} because {explanation}". Entailment maps to "Yes" and both neutral and contradiction map to "No", giving what the authors call multi-step verbalizing.
  • Ablations show that two things matter for the gains: the label verbalization itself and a genuine causal link between explanation and label.
  • For the T5-3B column pair, ANLI R3 rises from 57.9 to 74.8 and Counter-NLI RP&RH rises from 79.3 to 88.3. T5 variants did not improve on the HANS Subsequence and Constituent subsets.
  • In the PPMI analysis, the cue word "frowning" was linked to contradiction at about 0.1 in the label-only setting and dropped to near 0 with explanations.
  • The human evaluation used 100 random T0 outputs on Amazon Mechanical Turk, with three ratings each on a 5-point Likert scale. An unrelated "attention check" explanation was mixed in, and presentation order was shuffled.

Interesting ideas

  • I1. Breaking the label into a coarse binary decision followed by the fine label gives the model a staged commitment instead of one opaque token; staged outputs make errors easier to localize.
  • I2. PPMI between single input tokens and output labels serves as a direct diagnostic of shortcut learning, and comparing it before and after a training change shows whether cues were actually neutralized.
  • I3. Pairing explanations with labels they do not causally support isolates whether the content of the explanation matters or merely its presence.
  • I4. In crowd evaluation, an unrelated "attention check" explanation is inserted at a random position, and any HIT where it scores as high as the gold one is flagged.
  • I5. Hypothesis-only evaluation, which tests whether a label can be guessed without the premise, offers a simple way to expose dataset artifacts.

For Mova

How these ideas could serve Mova, our small open language model — its files, stages and tests:

  • I1 (answer): Readers answering yes/no or multiple-choice questions could first emit a coarse verdict (supported / not supported) and only then the fine category, logging each stage separately.
  • I2 (learn): Run a token–label PPMI scan over silver data and LLM-teacher outputs to detect words that predict answers or frame labels on their own, and use it as a negative-control check after retraining.
  • I3 (learn): When training components on teacher rationales, add controls with mismatched rationales to confirm the student uses the justification rather than its format.
  • I4 (answer): Evaluations of Mova's generated explanations for fable or school-test answers could include deliberately irrelevant distractor explanations, to verify that judges (human or LLM) are actually reading them.
  • I5 (answer): Feed question-only inputs (no story) to the QA readers; any above-chance accuracy indicates the test set or reader leaks answers through surface cues.

Our summary of the paper, in our own words — not its text. The work belongs to its authors; read it at the links above.

 Institutions

🎓Tohoku UniversitySendai · Japan20 works · 514 citations🔬RIKEN Center for Advanced Intelligence ProjectTokyo · Japan15 works · 957 citations🔬RIKENWako · Japan26 works · 526 citations

 Fields, topics and concepts

📚 Shelves 4The field of science this work belongs to (OpenAlex fields).Computer Science · 16,667 worksArtificial Intelligence · 16,194 worksArts and Humanities · 16,039 worksLanguage and Linguistics · 15,891 works

🧭 Topics 1Research subjects and languages the work is about.Reasoning · 1,426 works

Cite

Cite as Pride Kavumba, Ana Brassard, Benjamin Heinzerling and Kentaro Inui. 2023. Prompting for explanations improves Adversarial NLI. Is this true? \Yes\ it is \true\ because ıt weakens superficial cues\. Findings of the Association for Computational Linguistics: EACL 2023.

The paper itself, not this page. BibTeX from the library's record of the work. DOI 10.18653/v1/2023.findings-eacl.162

@inproceedings{kavumba2023prompting,
  title = {{Prompting for explanations improves Adversarial NLI. Is this true? Yes it is true because ıt weakens superficial cues}},
  author = {Pride Kavumba, Ana Brassard, Benjamin Heinzerling and Kentaro Inui},
  booktitle = {Findings of the Association for Computational Linguistics: EACL 2023},
  year = {2023},
  doi = {10.18653/v1/2023.findings-eacl.162},
  url = {https://doi.org/10.18653/v1/2023.findings-eacl.162}
}

References

50 references, 30 of them with a card of ours. ↗ leads to the work, «find» searches for one we have no id for yet.

  1. Stephen Bach, Victor Sanh, Zheng Xin Yong, Albert Webson, Colin Raffel, Nihal V. Nayak, Abheesht Sharma, Taewoon Kim, M Saiful Bari, Thibault Fevry, Zaid Alyafeai, Manan Dey, Andrea Santilli, Zhiqing Sun, Srulik Ben-david, Canwen Xu, Gunjan Chhablani, Han Wang, Jason Fries, Maged Al-shaibani, Shanya Sharma, Urmish Thakker, Khalid Almubarak, Xiangru Tang, Dragomir Radev, Mike Tian-jian Jiang, and Alexander Rush. 2022. Prompt-Source: An integrated development environment and repository for natural language prompts. In Proceedings of the 60th Annual Meeting of the Association for Computational Linguistics: System Demonstrations, pages 93–104, Dublin, Ireland. Association for Computational Linguistics. find
  2. Jasmijn Bastings, Wilker Aziz, and Ivan Titov. 2019. Interpretable neural predictions with differentiable binary variables. In Proceedings of the 57th Annual Meeting of the Association for Computational Linguistics, pages 2963–2977, Florence, Italy. Association for Computational Linguistics.
  3. Yonatan Belinkov, Adam Poliak, Stuart Shieber, Benjamin Van Durme, and Alexander Rush. 2019. On adversarial removal of hypothesis-only bias in natural language inference. In Proceedings of the Eighth Joint Conference on Lexical and Computational Semantics (*SEM 2019), pages 256–262, Minneapolis, Minnesota. Association for Computational Linguistics. ↗
  4. Samuel R. Bowman, Gabor Angeli, Christopher Potts, and Christopher D. Manning. 2015. A large annotated corpus for learning natural language inference. In Proceedings of the 2015 Conference on Empirical Methods in Natural Language Processing, pages 632–642, Lisbon, Portugal. Association for Computational Linguistics. ↗
  5. Ronan Le Bras, Swabha Swayamdipta, Chandra Bhagavatula, Rowan Zellers, Matthew Peters, Ashish Sabharwal, and Yejin Choi. 2020. Adversarial filters of dataset biases. In Proceedings of the 37th International Conference on Machine Learning, volume 119 of Proceedings of Machine Learning Research, pages 1078–1088. PMLR. ↗
  6. Tom Brown, Benjamin Mann, Nick Ryder, Melanie Subbiah, Jared D Kaplan, Prafulla Dhariwal, Arvind Neelakantan, Pranav Shyam, Girish Sastry, Amanda Askell, Sandhini Agarwal, Ariel Herbert-Voss, Gretchen Krueger, Tom Henighan, Rewon Child, Aditya Ramesh, Daniel Ziegler, Jeffrey Wu, Clemens Winter, Chris Hesse, Mark Chen, Eric Sigler, Mateusz Litwin, Scott Gray, Benjamin Chess, Jack Clark, Christopher Berner, Sam McCandlish, Alec Radford, Ilya Sutskever, and Dario Amodei. 2020. Language models are few-shot learners. In Advances in Neural Information Processing Systems, volume 33, pages 1877–1901. Curran Associates, Inc.
  7. Oana-Maria Camburu, Tim Rocktäschel, Thomas Lukasiewicz, and Phil Blunsom. 2018. e-snli: Natural language inference with natural language explanations. In S. Bengio, H. Wallach, H. Larochelle, K. Grauman, N. Cesa-Bianchi, and R. Garnett, editors, Advances in Neural Information Processing Systems 31, pages 9539–9549. Curran Associates, Inc. ↗
  8. Howard Chen, Jacqueline He, Karthik Narasimhan, and Danqi Chen. 2022. Can rationalization improve robustness? In Proceedings of the 2022 Conference of the North American Chapter of the Association for Computational Linguistics: Human Language Technologies, pages 3792–3805, Seattle, United States. Association for Computational Linguistics. ↗
  9. Jay DeYoung, Sarthak Jain, Nazneen Fatema Rajani, Eric Lehman, Caiming Xiong, Richard Socher, and Byron C. Wallace. 2020. ERASER: A benchmark to evaluate rationalized NLP models. In Proceedings of the 58th Annual Meeting of the Association for Computational Linguistics, pages 4443–4458, Online. Association for Computational Linguistics.
  10. Robert Geirhos, Jörn-Henrik Jacobsen, Claudio Michaelis, Richard Zemel, Wieland Brendel, Matthias Bethge, and Felix A. Wichmann. 2020. Shortcut learning in deep neural networks. Nature Machine Intelligence, 2(11):665–673. ↗
  11. Suchin Gururangan, Swabha Swayamdipta, Omer Levy, Roy Schwartz, Samuel Bowman, and Noah A. Smith. 2018a. Annotation artifacts in natural language inference data. In Proceedings of the 2018 Conference of the North American Chapter of the Association for Computational Linguistics: Human Language Technologies, Volume 2 (Short Papers), pages 107–112, New Orleans, Louisiana. Association for Computational Linguistics.
  12. Suchin Gururangan, Swabha Swayamdipta, Omer Levy, Roy Schwartz, Samuel Bowman, and Noah A. Smith. 2018b. Annotation artifacts in natural language inference data. In Proceedings of the 2018 Conference of the North American Chapter of the Association for Computational Linguistics: Human Language Technologies, Volume 2 (Short Papers), pages 107–112, New Orleans, Louisiana. Association for Computational Linguistics.
  13. Divyansh Kaushik, Eduard Hovy, and Zachary C Lipton. 2020. Learning the difference that makes a difference with counterfactually augmented data. International Conference on Learning Representations (ICLR). ↗
  14. Pride Kavumba, Naoya Inoue, Benjamin Heinzerling, Keshav Singh, Paul Reisert, and Kentaro Inui. 2019. When choosing plausible alternatives, clever hans can be clever. In Proceedings of the First Workshop on Commonsense Inference in Natural Language Processing, pages 33–42, Hong Kong, China. Association for Computational Linguistics. find
  15. Diederik P. Kingma and Jimmy Ba. 2015. Adam: A method for stochastic optimization. In 3rd International Conference on Learning Representations, ICLR 2015, San Diego, CA, USA, May 7-9, 2015, Conference Track Proceedings. ↗
  16. Tao Lei, Regina Barzilay, and Tommi Jaakkola. 2016. Rationalizing neural predictions. In Proceedings of the 2016 Conference on Empirical Methods in Natural Language Processing, pages 107–117, Austin, Texas. Association for Computational Linguistics.
  17. Mike Lewis, Yinhan Liu, Naman Goyal, Marjan Ghazvininejad, Abdelrahman Mohamed, Omer Levy, Veselin Stoyanov, and Luke Zettlemoyer. 2020. BART: Denoising sequence-to-sequence pre-training for natural language generation, translation, and comprehension. In Proceedings of the 58th Annual Meeting of the Association for Computational Linguistics, pages 7871–7880, Online. Association for Computational Linguistics. ↗
  18. Xiaodong Liu, Hao Cheng, Pengcheng He, Weizhu Chen, Yu Wang, Hoifung Poon, and Jianfeng Gao. 2020. Adversarial training for large neural language models. ↗
  19. Ana Marasovic, Iz Beltagy, Doug Downey, and Matthew Peters. 2022. Few-shot self-rationalization with natural language prompts. In Findings of the Association for Computational Linguistics: NAACL 2022, pages 410–424, Seattle, United States. Association for Computational Linguistics.
  20. Tom McCoy, Ellie Pavlick, and Tal Linzen. 2019. Right for the wrong reasons: Diagnosing syntactic heuristics in natural language inference. In Proceedings of the 57th Annual Meeting of the Association for Computational Linguistics, pages 3428–3448, Florence, Italy. Association for Computational Linguistics. ↗
  21. Sharan Narang, Colin Raffel, Katherine Lee, Adam Roberts, Noah Fiedel, and Karishma Malkan. 2020. Wt5?! training text-to-text models to explain their predictions. ↗
  22. Yixin Nie, Adina Williams, Emily Dinan, Mohit Bansal, Jason Weston, and Douwe Kiela. 2020. Adversarial NLI: A new benchmark for natural language understanding. In Proceedings of the 58th Annual Meeting of the Association for Computational Linguistics. Association for Computational Linguistics.
  23. Timothy Niven and Hung-Yu Kao. 2019. Probing neural network comprehension of natural language arguments. In Proceedings of the 57th Annual Meeting of the Association for Computational Linguistics, pages 4658–4664, Florence, Italy. Association for Computational Linguistics. find
  24. Adam Poliak, Jason Naradowsky, Aparajita Haldar, Rachel Rudinger, and Benjamin Van Durme. 2018. Hypothesis only baselines in natural language inference. In Proceedings of the Seventh Joint Conference on Lexical and Computational Semantics, pages 180–191, New Orleans, Louisiana. Association for Computational Linguistics. ↗
  25. Colin Raffel, Noam Shazeer, Adam Roberts, Katherine Lee, Sharan Narang, Michael Matena, Yanqi Zhou, Wei Li, and Peter J. Liu. 2020. Exploring the limits of transfer learning with a unified text-to-text transformer. Journal of Machine Learning Research, 21(140):1–67. ↗
  26. Samyam Rajbhandari, Jeff Rasley, Olatunji Ruwase, and Yuxiong He. 2020. Zero: Memory optimizations toward training trillion parameter models. find
  27. Nils Reimers and Iryna Gurevych. 2019. Sentence-BERT: Sentence embeddings using Siamese BERT-networks. In Proceedings of the 2019 Conference on Empirical Methods in Natural Language Processing and the 9th International Joint Conference on Natural Language Processing (EMNLP-IJCNLP), pages 3982–3992, Hong Kong, China. Association for Computational Linguistics. ↗
  28. Jie Ren, Samyam Rajbhandari, Reza Yazdani Aminabadi, Olatunji Ruwase, Shuangyan Yang, Minjia Zhang, Dong Li, and Yuxiong He. 2021. Zero-offload: Democratizing billion-scale model training. ↗
  29. Keisuke Sakaguchi, Ronan Le Bras, Chandra Bhagavatula, and Yejin Choi. 2020. Winogrande: An adversarial winograd schema challenge at scale. Proceedings of the AAAI Conference on Artificial Intelligence, 34(05):8732–8740. find
  30. Victor Sanh, Albert Webson, Colin Raffel, Stephen H. Bach, Lintang Sutawika, Zaid Alyafeai, Antoine Chaffin, Arnaud Stiegler, Teven Le Scao, Arun Raja, Manan Dey, M Saiful Bari, Canwen Xu, Urmish Thakker, Shanya Sharma Sharma, Eliza Szczechla, Taewoon Kim, Gunjan Chhablani, Nihal Nayak, Debajyoti Datta, Jonathan Chang, Mike Tian-Jian Jiang, Han Wang, Matteo Manica, Sheng Shen, Zheng Xin Yong, Harshit Pandey, Rachel Bawden, Thomas Wang, Trishala Neeraj, Jos Rozen, Abheesht Sharma, Andrea Santilli, Thibault Fevry, Jason Alan Fries, Ryan Teehan, Stella Biderman, Leo Gao, Tali Bers, Thomas Wolf, and Alexander M. Rush. 2021. Multi-task prompted training enables zero-shot task generalization.
  31. Timo Schick and Hinrich Schütze. 2021a. Exploiting cloze-questions for few-shot text classification and natural language inference. In Proceedings of the 16th Conference of the European Chapter of the Association for Computational Linguistics: Main Volume, pages 255–269, Online. Association for Computational Linguistics.
  32. Timo Schick and Hinrich Schütze. 2021b. It’s not just size that matters: Small language models are also few-shot learners. In Proceedings of the 2021 Conference of the North American Chapter of the Association for Computational Linguistics: Human Language Technologies, pages 2339–2352, Online. Association for Computational Linguistics.
  33. Tal Schuster, Darsh Shah, Yun Jie Serene Yeo, Daniel Roberto Filizzola Ortiz, Enrico Santus, and Regina Barzilay. 2019a. Towards debiasing fact verification models. In Proceedings of the 2019 Conference on Empirical Methods in Natural Language Processing and the 9th International Joint Conference on Natural Language Processing (EMNLP-IJCNLP), pages 3419–3425, Hong Kong, China. Association for Computational Linguistics.
  34. Tal Schuster, Darsh Shah, Yun Jie Serene Yeo, Daniel Roberto Filizzola Ortiz, Enrico Santus, and Regina Barzilay. 2019b. Towards debiasing fact verification models. In Proceedings of the 2019 Conference on Empirical Methods in Natural Language Processing and the 9th International Joint Conference on Natural Language Processing (EMNLP-IJCNLP), pages 3410–3416, Hong Kong, China. Association for Computational Linguistics.
  35. Megha Srivastava, Tatsunori Hashimoto, and Percy Liang. 2020. Robustness to spurious correlations via human annotations. In Proceedings of the 37th International Conference on Machine Learning, volume 119 of Proceedings of Machine Learning Research, pages 9109–9119. PMLR. strength in numbers: Avoiding the hypothesis-only bias in natural language inference via ensemble adversarial training. ↗
  36. Saku Sugawara, Kentaro Inui, Satoshi Sekine, and Akiko Aizawa. 2018. What makes reading comprehension questions easier? In Proceedings of the 2018 Conference on Empirical Methods in Natural Language Processing, pages 4208–4219, Brussels, Belgium. Association for Computational Linguistics.
  37. Alex Wang, Amanpreet Singh, Julian Michael, Felix Hill, Omer Levy, and Samuel Bowman. 2018. GLUE: A multi-task benchmark and analysis platform for natural language understanding. In Proceedings of the 2018 EMNLP Workshop BlackboxNLP: Analyzing and Interpreting Neural Networks for NLP, pages 353–355, Brussels, Belgium. Association for Computational Linguistics. ↗
  38. Boxin Wang, Shuohang Wang, Yu Cheng, Zhe Gan, Ruoxi Jia, Bo Li, and Jingjing Liu. 2021. Infobert: Improving robustness of language models from an information theoretic perspective. In International Conference on Learning Representations. ↗
  39. Dilin Wang, Chengyue Gong, and Qiang Liu. 2019a. Improving neural language modeling via adversarial training. In Proceedings of the 36th International Conference on Machine Learning, volume 97 of Proceedings of Machine Learning Research, pages 6555– 6565. PMLR. ↗
  40. Haohan Wang, Songwei Ge, Zachary C. Lipton, and Eric P. Xing. 2019b. Learning robust global representations by penalizing local predictive power. In Advances in Neural Information Processing Systems 32: Annual Conference on Neural Information Processing Systems 2019, NeurIPS 2019, 8-14 December 2019, Vancouver, BC, Canada, pages 10506–10518. find
  41. Haohan Wang, Zexue He, Zachary C. Lipton, and Eric P. Xing. 2019c. Learning robust representations by projecting superficial statistics out. In 7th International Conference on Learning Representations, ICLR 2019, New Orleans, LA, USA, May 6-9, 2019. OpenReview.net. ↗
  42. Zhao Wang and Aron Culotta. 2021. Robustness to spurious correlations in text classification via automatically generated counterfactuals. Proceedings of the AAAI Conference on Artificial Intelligence, 35(16):14024–14031. ↗
  43. Joe Stacey, Yonatan Belinkov, and Marek Rei. 2021. Supervising model attention with human explanations for robust natural language inference. find
  44. Sarah Wiegreffe, Jack Hessel, Swabha Swayamdipta, Mark Riedl, and Yejin Choi. 2022. Reframing human-AI collaboration for generating free-text explanations. In Proceedings of the 2022 Conference of the North American Chapter of the Association for Computational Linguistics: Human Language Technologies, pages 632–658, Seattle, United States. Association for Computational Linguistics. ↗
  45. Joe Stacey, Pasquale Minervini, Haim Dubossarsky, Sebastian Riedel, and Tim Rocktäschel. 2020. There is Sarah Wiegreffe, Ana Marasović, and Noah A. Smith. 2021. Measuring association between labels and free-text rationales. In Proceedings of the 2021 Conference on Empirical Methods in Natural Language Processing, pages 10266–10284, Online and Punta Cana, Dominican Republic. Association for Computational Linguistics.
  46. Thomas Wolf, Lysandre Debut, Victor Sanh, Julien Chaumond, Clement Delangue, Anthony Moi, Pierric Cistac, Tim Rault, R’emi Louf, Morgan Funtowicz, and Jamie Brew. 2019. Huggingface’s transformers: State-of-the-art natural language processing. ArXiv, abs/1910.03771. ↗
  47. Rowan Zellers, Yonatan Bisk, Roy Schwartz, and Yejin Choi. 2018. SWAG: A large-scale adversarial dataset for grounded commonsense inference. In Proceedings of the 2018 Conference on Empirical Methods in Natural Language Processing, pages 93–104, Brussels, Belgium. Association for Computational Linguistics. ↗
  48. Rowan Zellers, Ari Holtzman, Yonatan Bisk, Ali Farhadi, and Yejin Choi. 2019. HellaSwag: Can a machine really finish your sentence? In Proceedings of the 57th Annual Meeting of the Association for Computational Linguistics, pages 4791–4800, Florence, Italy. Association for Computational Linguistics. ↗
  49. Chen Zhu, Yu Cheng, Zhe Gan, Siqi Sun, Tom Goldstein, and Jingjing Liu. 2020. Freelb: Enhanced adversarial training for natural language understanding. In International Conference on Learning Representations. ↗
  50. Yukun Zhu, Ryan Kiros, Rich Zemel, Ruslan Salakhutdinov, Raquel Urtasun, Antonio Torralba, and Sanja Fidler. 2015. Aligning books and movies: Towards story-like visual explanations by watching movies and reading books. In The IEEE International Conference on Computer Vision (ICCV).

Updated

↑ back to top

This page in Markdown → https://md.research.elamur.ai/free/en/papers/kavumba-2023-prompting-explanations-improves-adversarial-nli/