Elamur Research⌕
☰
home / papers / 2023 / When Truth Matters - Addressing Pragmatic Categories in Natural Language Inference (NLI) by Large Language Models (LLMs)

When Truth Matters - Addressing Pragmatic Categories in Natural Language Inference (NLI) by Large Language Models (LLMs)

Summary

The paper argues that logical inference only applies to genuine claims. Questions, commands and incomplete fragments assert nothing that could be true or false, so no hypothesis can follow from them or contradict them. The authors give two conditions for a premise to express a determinate claim: it must have assertive force, and it must be sensible, meaning it identifies a subject and says something about it. They show that MNLI, the dominant NLI training corpus, includes such non-assertive premises with entailment or contradiction labels, and that models fine-tuned on it ignore the distinction. Using the symbolic semantic parser GKR, they detect non-assertive premises, relabel these pairs as neutral, and build several fine-tuning sets. They also release an expert-checked evaluation set. Models trained this way learn to treat non-assertive input correctly while losing only modest accuracy on MNLI-matched. A small manual probe of ChatGPT shows that it can state the textbook definition of entailment but often misapplies it. The work matters for claim verification, legal testimony analysis, educational feedback and chatbots, where claims must be separated from non-claims.

Key points

  • The paper proposes two criteria for premises that can enter inference: C1 requires assertive pragmatic force, and C2 requires a sensible assertion with a clear subject and predicate. The framework builds on speech act theory (Austin, Searle) and separates inference from presupposition.
  • GKR, used to flag non-assertive premises, reached an F-score of 84.3% in the authors' evaluation. As a result, the fine-tuning data contains some noise, but the manually verified evaluation set does not.
  • The released resources are one expert-curated gold-standard evaluation dataset and 7 fine-tuning datasets, available at github.com/retoj/whentruthmatters.
  • ChatGPT was tested on 96 MNLI pairs that GKR marked as non-assertive and that were originally intended as entailments. It still affirmed entailment in 54% of cases. It handled fragmentary premises reasonably well but did poorly on questions and commands.
  • The authors name two main limitations: the method depends on the parser's quality, and the ChatGPT sample is small and needs a larger follow-up study.

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

Fields: Linguistics & languages

Cite

The paper itself, not this page. BibTeX from the library's record of the work. DOI 10.18653/v1/2023.starsem-1.4

@inproceedings{gubelmann2023when,
  title = {{When Truth Matters - Addressing Pragmatic Categories in Natural Language Inference (NLI) by Large Language Models (LLMs)}},
  author = {Gubelmann, Reto and Kalouli, Aikaterini-lida and Niklaus, Christina and Handschuh, Siegfried},
  booktitle = {Proceedings of the 12th Joint Conference on Lexical and Computational Semantics (*SEM 2023)},
  year = {2023},
  doi = {10.18653/v1/2023.starsem-1.4},
  url = {https://doi.org/10.18653/v1/2023.starsem-1.4}
}

References

38 references, 6 of them with a card of ours. ↗ leads to the work, «find» searches for one we have no id for yet.

  1. Herbert P Grice. 1975. Logic and conversation. In Speech Acts, pages 41–58. Brill. ↗
  2. Kevin D Ashley. 2018. Automatically extracting meaning from legal texts: opportunities and challenges. Ga. St. UL Rev., 35:1117. find
  3. John Austin. 1962. Clarendon Press. find
  4. Jeroen Antonius Gerardus Groenendijk and Martin Johan Bastiaan Stokhof. 1984. Studies on the Semantics of Questions and the Pragmatics of Answers. Ph.D. thesis, Univ. Amsterdam. How to do things with words. find
  5. Reto Gubelmann. 2023. A Loosely Wittgensteinian Conception of the Linguistic Understanding of Artificial Neural Networks. Forthcoming in Grazer Philosophische Studien. find
  6. John Langshaw Austin. 1975. How to do things with words. Oxford university press. ↗
  7. Yejin Bang, Samuel Cahyawijaya, Nayeon Lee, Wenliang Dai, Dan Su, Bryan Wilie, Holy Lovenia, Ziwei Ji, Tiezheng Yu, Willy Chung, et al. 2023. A multi-task, multilingual, multimodal evaluation of chatgpt on reasoning, hallucination, and interactivity. arXiv preprint arXiv:2302.04023. ↗ our card
  8. Reto Gubelmann, Christina Niklaus, and Siegfried Handschuh. 2022. A philosophically-informed contribution to the generalization problem of neural natural language inference: Shallow heuristics, bias, and the varieties of inference. In Proceedings of the 3rd Natural Logic Meets Machine Learning Workshop (NALOMA III), pages 38–50. find
  9. Yinhan Liu, Myle Ott, Naman Goyal, Jingfei Du, Mandar Joshi, Danqi Chen, Omer Levy, Mike Lewis, Luke Zettlemoyer, and Veselin Stoyanov. 2019. Roberta: A robustly optimized bert pretraining approach. arXiv preprint arXiv:1907.11692. ↗
  10. Suchin Gururangan, Swabha Swayamdipta, Omer Levy, Roy Schwartz, Samuel Bowman, and Noah A Smith. 2018. 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. find
  11. Bill MacCartney. 2009. Natural Language Inference. Ph.D. thesis, Stanford University, Stanford, CA, USA. AAI3364139. find
  12. John Morris, Eli Lifland, Jin Yong Yoo, Jake Grigsby, Di Jin, and Yanjun Qi. 2020. Textattack: A framework for adversarial attacks, data augmentation, and adversarial training in nlp. In Proceedings of the 2020 Conference on Empirical Methods in Natural Language Processing: System Demonstrations, pages 119–126. find our card
  13. Pengcheng He, Xiaodong Liu, Jianfeng Gao, and Weizhu Chen. 2020. Deberta: Decoding-enhanced bert with disentangled attention. In International Conference on Learning Representations. ↗ our card
  14. Nan-Jiang Jiang and Marie-Catherine de Marneffe. 2022. Investigating reasons for disagreement in natural language inference. arXiv preprint arXiv:2209.03392. OpenAI. ChatGPT. ↗ our card
  15. Ellie Pavlick and Tom Kwiatkowski. 2019. Inherent disagreements in human textual inferences. Transactions of the Association for Computational Linguistics, 7:677–694. ↗ our card
  16. Aikaterini-Lida Kalouli. 2021. Hy-NLI : a Hybrid system for state-of-the-art Natural Language Inference. Ph.D. thesis, Universität Konstanz, Konstanz. find
  17. Aikaterini-Lida Kalouli, Annebeth Buis, Livy Real, Martha Palmer, and Valeria de Paiva. 2019. Explaining simple natural language inference. In Proceedings of the 13th Linguistic Annotation Workshop, pages 132–143, Florence, Italy. Association for Computational Linguistics. find
  18. Stanley Peters. 1979. A truth-conditional formulation of Karttunen’s account of presupposition. Synthese. An International Journal for Epistemology, Methodology and Philosophy of Science, 40:301–316. ↗
  19. Alvin Plantinga. 1974. The Nature of Necessity. Oxford: Oxford University Press. find
  20. Aikaterini-Lida Kalouli and Richard Crouch. 2018. Gkr: the graphical knowledge representation for semantic parsing. In Workshop on Computational Semantics beyond Events and Roles (SemBEaR 2018), pages 27–37. find
  21. 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. Association for Computational Linguistics. find
  22. Aikaterini-Lida Kalouli, Hai Hu, Alexander F. Webb, Lawrence S. Moss, and Valeria de Paiva. 2023. Curing the SICK and Other NLI Maladies. Computational Linguistics, pages 1–45. ↗
  23. Christopher Potts. 2004. The Logic of Conventional Implicatures, volume 7. OUP Oxford. find
  24. Lauri Karttunen and Stanley Peters. 1979. Conventional lmplicature. In Presupposition, pages 1–56. Brill. find
  25. Christopher Potts. 2015. Presupposition and implicature. The handbook of contemporary semantic theory, pages 168–202. ↗
  26. Zhenzhong Lan, Mingda Chen, Sebastian Goodman, Kevin Gimpel, Piyush Sharma, and Radu Soricut. 2019. Albert: A lite bert for self-supervised learning of language representations. In International Conference on Learning Representations. find
  27. Colin Raffel, Noam Shazeer, Adam Roberts, Katherine Lee, Sharan Narang, Michael Matena, Yanqi Zhou, Wei Li, and Peter J Liu. 2019. Exploring the limits of transfer learning with a unified text-to-text transformer. arXiv preprint arXiv:1910.10683. ↗
  28. Ernie Lepore and Matthew Stone. 2010. Against metaphorical meaning. Topoi. Orient – Occident, 29:165–180. ↗
  29. Chrysi Rapanta, Merce Garcia-Mila, and Sandra Gi-labert. 2013. What is meant by argumentative competence? an integrative review of methods of analysis and assessment in education. Review of Educational Research, 83(4):483–520. ↗
  30. Stephen C Levinson. 2017. Speech acts. In Oxford handbook of pragmatics, pages 199–216. Oxford: Oxford University Press. find
  31. 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. ArXiv, abs/1910.13461. ↗ our card
  32. Nils Reimers and Iryna Gurevych. 2019. Sentence-bert: Sentence embeddings using siamese bert-networks. arXiv preprint arXiv:1908.10084. ↗
  33. Bertrand Russell. 1905. 14(56):479–493. On denoting. find
  34. Mind, A Victor Sanh, Lysandre Debut, Julien Chaumond, and Thomas Wolf. 2019. Distilbert, a distilled version of bert: Smaller, faster, cheaper and lighter. arXiv preprint arXiv:1910.01108. ↗
  35. Williams et al. (2018, 1114) specify the following tasks for the crowdworkers: “This task will involve reading a line from a non-fiction article and writing three sentences that relate to it. The line will describe a situation or event. Using only this description and what you know about the world: John Searle. 1969. Speech Acts: An Essay in the Philosophy of Language. Cambridge: Cambridge University Press. find
  36. John Searle. 1985. Expression and meaning: Studies in the theory of speech acts. Cambridge: Cambridge University Press. • Write one sentence that is definitely correct about the situation or event in the line. find
  37. Stewart Shapiro and Teresa Kouri Kissel. 2021. Classical Logic. In Edward N. Zalta, editor, The Stanford Encyclopedia of Philosophy, Spring 2021 edition. Metaphysics Research Lab, Stanford University. find
  38. Raymond M. Smullyan. 1968. Dover. Full Instructions Given to Crowdworkers • Write one sentence that might be correct about the situation or event in the line. First-Order Logic. • Write one sentence that is definitely incorrect about the situation or event in the line. " Ashish Vaswani, Noam Shazeer, Niki Parmar, Jakob Uszkoreit, Llion Jones, Aidan N. Gomez, undefine-dukasz Kaiser, and Illia Polosukhin. 2017. Attention is all you need. In Proceedings of the 31st International Conference on Neural Information Processing Systems, NIPS’17, page 6000–6010, Red Hook, NY, USA. Curran Associates Inc. find

↑ back to top

This page in Markdown → https://md.research.elamur.ai/free/papers/gubelmann-2023-when-truth-matters-addressing-pragmatic/