Elamur Research⌕
☰
home / papers / 2022 / Distributed NLI: Learning to Predict Human Opinion Distributions for Language Reasoning

Distributed NLI: Learning to Predict Human Opinion Distributions for Language Reasoning

Paper · 2022 · Xiang Zhou, Yixin Nie and Mohit Bansal

Summary

The paper treats NLI as predicting how a crowd splits its judgements, not as picking one correct label. The authors call this task distributed NLI. They build on earlier datasets with many annotations per example, chiefly ChaosNLI-S and ChaosNLI-M (drawn from SNLI and MNLI), and add an experimental protocol for distribution labels. The model's output is a categorical distribution over entailment, neutral and contradiction, compared against the human distribution with KL and JSD. Instead of designing new architectures, they apply four estimation methods to strong pretrained classifiers: MC Dropout, Deep Ensemble, Re-Calibration (temperature scaling) and Distribution Distillation. All four beat the plain softmax baseline. MC Dropout works reasonably well without any distribution annotations. Re-Calibration gains further when some multi-annotated data is available. Both transfer passably to an out-of-domain test set. Even so, the best systems remain far below an estimated human upper bound. The work matters because ambiguous items have no single defensible label, and downstream users, including decision-support systems, need to see minority readings rather than only the majority vote.

Key points

  • Distributed NLI was evaluated with four approaches: Deep Ensemble, MC Dropout, Distribution Distillation, and Re-Calibration. The distillation variant uses Re-Calibration as its teacher.
  • The data comes from ChaosNLI (ChaosNLI-S, ChaosNLI-M), which has about 100 votes per item. One example splits E 57 / N 42 / C 1 over whether "a legend" means "a well known hero".
  • Re-Calibration's temperature was tuned only on KL. The temperature that is best for KL is not the best one for JSD, so the two metrics can disagree.
  • Error analysis on an MNLI item shows two different failures. The human split was 0.33 / 0.66 / 0.01. MC Dropout was overconfident on neutral (about 0.84). Re-Calibration was smoother but put about 0.10 on an implausible contradiction.
  • The authors give guidelines for choosing a method depending on how much distribution-annotated data exists. They also warn against using current NLI models in high-stakes decisions.

Interesting ideas

  • I1. Treat the gold answer as a vote distribution rather than one label, since some items really are ambiguous and forcing one label misrepresents them.
  • I2. MC Dropout, which runs a trained model several times with dropout left on, recovers a useful disagreement estimate without any multi-annotated data.
  • I3. A single temperature fitted on a small multi-annotated set noticeably improves how closely predicted spreads match human ones, at almost no cost.
  • I4. The two methods fail in opposite ways: one is overconfident, the other smooths probability onto absurd labels. A comparison should therefore inspect both failure types, not only average KL.
  • I5. Disagreement often comes from pragmatic inference, for example readers assuming a clerkship was beneficial when the text never says so.

For Mova

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

  • I1 (answer): School-test and fable QA readers could return a graded spread over candidate answers on ambiguous questions and be scored against that spread rather than a single key.
  • I2 (tree): gparse could sample several dropout passes per sentence and report per-arc attachment variance, flagging genuinely ambiguous PP attachments instead of committing silently.
  • I3 (learn): a small set where several witness parsers or LLM teachers voted could be used to calibrate confidence scores before they gate silver-data acceptance.
  • I4 (learn): evaluation with negative controls could explicitly check that smoothing never raises the probability of logically excluded readings, such as contradiction on a pragmatic-entailment case.
  • I5 (frames): implicature in prag could carry a strength or likelihood instead of a yes/no flag, so the world model can hold an inference as probable rather than asserted.

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

🎓University of North Carolina at Chapel HillChapel Hill · United States37 works · 14,332 citations

📚 Shelves 4The field of science this work belongs to (OpenAlex fields).Computer Science · 9,086 worksArtificial Intelligence · 8,889 worksArts and Humanities · 9,296 worksLanguage and Linguistics · 9,206 works

🧭 Topics 1Research subjects and languages the work is about.Reasoning · 510 works

Cite

Cite as Xiang Zhou, Yixin Nie and Mohit Bansal. 2022. Distributed NLI: Learning to Predict Human Opinion Distributions for Language Reasoning. Findings of the Association for Computational Linguistics: ACL 2022.

The paper itself, not this page. BibTeX from the library's record of the work. DOI 10.18653/v1/2022.findings-acl.79

@inproceedings{zhou2022distributed,
  title = {{Distributed NLI: Learning to Predict Human Opinion Distributions for Language Reasoning}},
  author = {Xiang Zhou, Yixin Nie and Mohit Bansal},
  booktitle = {Findings of the Association for Computational Linguistics: ACL 2022},
  year = {2022},
  doi = {10.18653/v1/2022.findings-acl.79},
  url = {https://doi.org/10.18653/v1/2022.findings-acl.79}
}

References

67 references, 25 of them with a card of ours. ↗ leads to the work, «find» searches for one we have no id for yet.

  1. John Armour and Mari Sako. 2020. Ai-enabled business models in legal services: from traditional law firms to next-generation law companies? Journal of Professions and Organization, 7(1):27–46. ↗
  2. Chandra Bhagavatula, Ronan Le Bras, Chaitanya Malaviya, Keisuke Sakaguchi, Ari Holtzman, Hannah Rashkin, Doug Downey, Wen-tau Yih, and Yejin Choi. 2020. Abductive commonsense reasoning. In 8th International Conference on Learning Representations, ICLR 2020, Addis Ababa, Ethiopia, April 26-30, 2020. OpenReview.net. ↗
  3. Samuel 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. challenge. In Machine Learning Challenges Workshop, pages 177–190. Springer. ↗
  4. Alexander Philip Dawid and Allan M Skene. 1979. Maximum likelihood estimation of observer error-rates using the em algorithm. Journal of the Royal Statistical Society: Series C (Applied Statistics), 28(1):20–28. ↗
  5. Shrey Desai and Greg Durrett. 2020. Calibration of pre-trained transformers. In Proceedings of the 2020 Conference on Empirical Methods in Natural Language Processing (EMNLP), pages 295–302. find
  6. Anca Dumitrache, Lora Aroyo, and Chris Welty. 2019. A crowdsourced frame disambiguation corpus with ambiguity. In Proceedings of the 2019 Conference of the North American Chapter of the Association for Computational Linguistics: Human Language Technologies, Volume 1 (Long and Short Papers), pages 2164–2170. find
  7. William D. Eggers, David Schatsky, and Peter Viechnicki. 2017. Ai-augmented government using cognitive technologies to redesign public sector work. find
  8. Tobias Falke, Leonardo F. R. Ribeiro, Prasetya Ajie Utama, Ido Dagan, and Iryna Gurevych. 2019. Ranking generated summaries by correctness: An interesting but challenging application for natural language inference. In Proceedings of the 57th Annual Meeting of the Association for Computational Linguistics, pages 2214–2220, Florence, Italy. Association for Computational Linguistics.
  9. Tommaso Fornaciari, Alexandra Uma, Silviu Paun, Barbara Plank, Dirk Hovy, and Massimo Poesio. 2021. Beyond black & white: Leveraging annotator disagreement via soft-label multi-task learning. In Proceedings of the 2021 Conference of the North American Chapter of the Association for Computational Linguistics: Human Language Technologies, Volume 2 (Short Papers).
  10. Yarin Gal. 2016. Uncertainty in Deep Learning. Ph.D. thesis, University of Cambridge. find
  11. Alexander Braylan and Matthew Lease. 2020. Modeling and aggregation of complex annotations via annotation distances. In Proceedings of The Web Conference 2020, pages 1807–1818. find
  12. Tongfei Chen, Zhengping Jiang, Adam Poliak, Keisuke Sakaguchi, and Benjamin Van Durme. 2020. Uncertain natural language inference. In Proceedings of the 58th Annual Meeting of the Association for Computational Linguistics, pages 8772–8779, Online. Association for Computational Linguistics. ↗
  13. Ido Dagan, Oren Glickman, and Bernardo Magnini. 2005. The pascal recognising textual entailment Yarin Gal and Zoubin Ghahramani. 2016. Dropout as a bayesian approximation: Representing model uncertainty in deep learning. In Proceedings of The 33rd International Conference on Machine Learning, volume 48 of Proceedings of Machine Learning Research, pages 1050–1059, New York, New York, USA. PMLR. find
  14. William Gantt, Benjamin Kane, and Aaron Steven White. 2020. Natural language inference with mixed effects. In Proceedings of the Ninth Joint Conference on Lexical and Computational Semantics, pages 81–87, Barcelona, Spain (Online). Association for Computational Linguistics. ↗
  15. Ramanathan Gnanadesikan and Martin B Wilk. 1968. Probability plotting methods for the analysis of data. Biometrika, 55(1):1–17. ↗
  16. Mitchell L Gordon, Kaitlyn Zhou, Kayur Patel, Tatsunori Hashimoto, and Michael S Bernstein. 2021. The disagreement deconvolution: Bringing machine learning performance metrics in line with reality. In Proceedings of the 2021 CHI Conference on Human Factors in Computing Systems, pages 1–14. find
  17. Chuan Guo, Geoff Pleiss, Yu Sun, and Kilian Q Weinberger. 2017. On calibration of modern neural networks. In Proceedings of the 34th International Conference on Machine Learning-Volume 70, pages 1321–1330. ↗
  18. Geoffrey Hinton, Oriol Vinyals, and Jeff Dean. 2015. Distilling the knowledge in a neural network. arXiv preprint arXiv:1503.02531. ↗
  19. Christopher D. Manning. 2006. Local textual inference: It’s hard to circumscribe. find
  20. Johannes Mario Meissner, Napat Thumwanit, Saku Sugawara, and Akiko Aizawa. 2021. Embracing ambiguity: Shifting the training target of NLI models. In Proceedings of the 59th Annual Meeting of the Association for Computational Linguistics and the 11th International Joint Conference on Natural Language Processing (Volume 2: Short Papers), pages 862–869, Online. Association for Computational Linguistics. ↗
  21. Nicholas Metropolis and Stanislaw Ulam. 1949. The monte carlo method. Journal of the American statistical association, 44(247):335–341. ↗
  22. Radford M Neal. 1995. BAYESIAN LEARNING FOR NEURAL NETWORKS. Ph.D. thesis, University of Toronto. ↗
  23. Dirk Hovy, Taylor Berg-Kirkpatrick, Ashish Vaswani, and Eduard Hovy. 2013. Learning whom to trust with mace. In Proceedings of the 2013 Conference of the North American Chapter of the Association for Computational Linguistics: Human Language Technologies, pages 1120–1130.
  24. Grace Newman, Ryan Frazier, and Kandace Miller. 2020. The 2020 recruiting trends report: Covid-19 revision. find
  25. Edwin T Jaynes. 2003. Probability theory: The logic of science. Cambridge university press. find
  26. Alex Kendall and Yarin Gal. 2017. What uncertainties do we need in bayesian deep learning for computer vision? In Proceedings of the 31st International Conference on Neural Information Processing Systems, pages 5580–5590. ↗
  27. Ágnes Melinda Kovács, Ernő Téglás, and Ansgar Denis Endress. 2010. The social sense: Susceptibility to others’ beliefs in human infants and adults. Science, 330(6012):1830–1834. ↗
  28. Khanh Nguyen and Brendan O’Connor. 2015. Posterior calibration and exploratory analysis for natural language processing models. In Proceedings of the 2015 Conference on Empirical Methods in Natural Language Processing, pages 1587–1598.
  29. Yixin Nie and Mohit Bansal. 2017. Shortcut-stacked sentence encoders for multi-domain inference. In Proceedings of the 2nd Workshop on Evaluating Vector Space Representations for NLP, pages 41–45. find
  30. Aviral Kumar and Sunita Sarawagi. 2019. Calibration of encoder decoder models for neural machine translation. arXiv preprint arXiv:1903.00802. ↗
  31. Yixin Nie, Adina Williams, Emily Dinan, Mohit Bansal, Jason Weston, and Douwe Kiela. 2020a. 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. find
  32. Balaji Lakshminarayanan, Alexander Pritzel, and Charles Blundell. 2017. Simple and scalable predictive uncertainty estimation using deep ensembles. In Proceedings of the 31st International Conference on Neural Information Processing Systems, pages 6405–6416. ↗
  33. Yixin Nie, Xiang Zhou, and Mohit Bansal. 2020b. What can we learn from collective human opinions on natural language inference data? In Proceedings of the 2020 Conference on Empirical Methods in Natural Language Processing (EMNLP), pages 9131–9143.
  34. Maolin Li, Hiroya Takamura, and Sophia Ananiadou. 2020. A neural model for aggregating coreference annotation in crowdsourcing. In Proceedings of the 28th International Conference on Computational Linguistics, pages 5760–5773. find
  35. 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. ↗
  36. David JC MacKay. 1992. A practical bayesian framework for backpropagation networks. Neural computation, 4(3):448–472. ↗
  37. Jessica Ochmann and S. Laumer. 2020. Ai recruitment: Explaining job seekers’ acceptance of automation in human resource management. In Wirtschaftsinformatik. ↗
  38. Myle Ott, Michael Auli, David Grangier, and Marc’Aurelio Ranzato. 2018. Analyzing uncertainty in neural machine translation. In International Conference on Machine Learning, pages 3956–3965. PMLR. ↗
  39. Ankur Parikh, Oscar Täckström, Dipanjan Das, and Jakob Uszkoreit. 2016. A decomposable attention model for natural language inference. In Proceedings of the 2016 Conference on Empirical Methods in Natural Language Processing, pages 2249–2255. ↗
  40. Ramakanth Pasunuru and Mohit Bansal. 2017. Multi-task video captioning with video and entailment generation. In Proceedings of the 55th Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers), pages 1273–1283, Vancouver, Canada. Association for Computational Linguistics. find
  41. Silviu Paun, Jon Chamberlain, Udo Kruschwitz, Juntao Yu, and Massimo Poesio. 2018. A probabilistic annotation model for crowdsourcing coreference. In Proceedings of the 2018 Conference on Empirical Methods in Natural Language Processing, pages 1926–1937.
  42. Ellie Pavlick and Tom Kwiatkowski. 2019. Inherent disagreements in human textual inferences. Transactions of the Association for Computational Linguistics, 7:677–694. ↗
  43. Joshua C Peterson, Ruairidh M Battleday, Thomas L Griffiths, and Olga Russakovsky. 2019. Human uncertainty makes classification more robust. In Proceedings of the IEEE/CVF International Conference on Computer Vision, pages 9617–9626. find
  44. Barbara Plank, Dirk Hovy, and Anders Søgaard. 2014. Learning part-of-speech taggers with inter-annotator agreement loss. In Proceedings of the 14th Conference of the European Chapter of the Association for Computational Linguistics, pages 742–751.
  45. Barbara Plank, Anders Søgaard, and Yoav Goldberg. 2016. Multilingual part-of-speech tagging with bidirectional long short-term memory models and auxiliary loss. In Proceedings of the 54th Annual Meeting of the Association for Computational Linguistics (Volume 2: Short Papers), pages 412–418.
  46. Massimo Poesio and Ron Artstein. 2005. The reliability of anaphoric annotation, reconsidered: Taking ambiguity into account. In Proceedings of the workshop on frontiers in corpus annotations ii: Pie in the sky, pages 76–83.
  47. Massimo Poesio, Jon Chamberlain, Silviu Paun, Juntao Yu, Alexandra Uma, and Udo Kruschwitz. 2019. A crowdsourced corpus of multiple judgments and disagreement on anaphoric interpretation. In Proceedings of the 2019 Conference of the North American Chapter of the Association for Computational Linguistics: Human Language Technologies, Volume 1 (Long and Short Papers), pages 1778–1789. ↗
  48. Massimo Poesio, Uwe Reyle, and Rosemary Stevenson. 2008. Justified sloppiness in anaphoric reference. In Computing meaning, pages 11–31. Springer. ↗
  49. Adam Poliak, Aparajita Haldar, Rachel Rudinger, J Edward Hu, Ellie Pavlick, Aaron Steven White, and Benjamin Van Durme. 2018. Collecting diverse natural language inference problems for sentence representation evaluation. In Proceedings of the 2018 Conference on Empirical Methods in Natural Language Processing. ↗
  50. Christopher Potts, Daniel Lassiter, Roger Levy, and Michael C Frank. 2016. Embedded implicatures as pragmatic inferences under compositional lexical uncertainty. Journal of Semantics, 33(4):755–802. ↗
  51. Marco Tulio Ribeiro, Sameer Singh, and Carlos Guestrin. 2016. " why should i trust you?" explaining the predictions of any classifier. In Proceedings of the 22nd ACM SIGKDD international conference on knowledge discovery and data mining, pages 1135–1144.
  52. Filipe Rodrigues, Mariana Lourenco, Bernardete Ribeiro, and Francisco C Pereira. 2017. Learning supervised topic models for classification and regression from crowds. IEEE transactions on pattern analysis and machine intelligence, 39(12):2409– 2422. ↗
  53. Edwin Simpson, Erik-Lân Do Dinh, Tristan Miller, and Iryna Gurevych. 2019. Predicting humorousness and metaphor novelty with gaussian process preference learning. In Proceedings of the 57th Annual Meeting of the Association for Computational Linguistics, pages 5716–5728. find
  54. Richard Socher, Alex Perelygin, Jean Wu, Jason Chuang, Christopher D Manning, Andrew Y Ng, and Christopher Potts. 2013. Recursive deep models for semantic compositionality over a sentiment treebank. In Proceedings of the 2013 conference on empirical methods in natural language processing, pages 1631–1642. find
  55. Nitish Srivastava, Geoffrey Hinton, Alex Krizhevsky, Ilya Sutskever, and Ruslan Salakhutdinov. 2014. Dropout: a simple way to prevent neural networks from overfitting. The journal of machine learning research, 15(1):1929–1958. find
  56. Harry Surden. 2019. Artificial intelligence and law: An overview. Georgia State University Law Review, 35:19–22. find
  57. Alon Talmor, Jonathan Herzig, Nicholas Lourie, and Jonathan Berant. 2019. CommonsenseQA: A question answering challenge targeting commonsense knowledge. In Proceedings of the 2019 Conference of the North American Chapter of the Association for Computational Linguistics: Human Language Technologies, Volume 1 (Long and Short Papers), pages 4149–4158, Minneapolis, Minnesota. Association for Computational Linguistics. ↗
  58. Alex Wang, Amanpreet Singh, Julian Michael, Felix Hill, Omer Levy, and Samuel R. Bowman. 2019. GLUE: A multi-task benchmark and analysis platform for natural language understanding. In 7th International Conference on Learning Representations, ICLR 2019, New Orleans, LA, USA, May 6-9, 2019. OpenReview.net. ↗
  59. Shuo Wang, Zhaopeng Tu, Shuming Shi, and Yang Liu. 2020. On the inference calibration of neural machine translation. In Proceedings of the 58th Annual Meeting of the Association for Computational Linguistics, pages 3070–3079. find
  60. Adina Williams, Nikita Nangia, and Samuel Bowman. 2018. A broad-coverage challenge corpus for sentence understanding through inference. In Proceedings of the 2018 Conference of the North American Chapter of the Association for Computational Linguistics: Human Language Technologies, Volume 1 (Long Papers), pages 1112–1122. find
  61. Yijun Xiao and William Yang Wang. 2019. Quantifying uncertainties in natural language processing tasks. In Proceedings of the AAAI Conference on Artificial Intelligence, volume 33, pages 7322–7329. find
  62. Jiacheng Xu, Shrey Desai, and Greg Durrett. 2020. Understanding neural abstractive summarization models via uncertainty. In Proceedings of the 2020 Conference on Empirical Methods in Natural Language Processing (EMNLP), pages 6275–6281. find
  63. Daniel Zeman. 2010. Hard problems of tagset conversion. In Proceedings of the Second International Conference on Global Interoperability for Language Resources, pages 181–185. find
  64. Sheng Zhang, Rachel Rudinger, Kevin Duh, and Benjamin Van Durme. 2017. Ordinal common-sense inference. Transactions of the Association for Computational Linguistics, 5:379–395. ↗
  65. Shujian Zhang, Chengyue Gong, and Eunsol Choi. 2021a. Capturing label distribution: A case study in nli. arXiv preprint arXiv:2102.06859. ↗
  66. Shujian Zhang, Chengyue Gong, and Eunsol Choi. 2021b. Learning from uneven training data: Unlabeled, single label, and multiple labels. In Proceedings of the 2021 Conference on Empirical Methods in Natural Language Processing (EMNLP). find
  67. Xinliang Frederick Zhang and Marie-Catherine de Marneffe. 2021. Identifying inherent disagreement in natural language inference. In Proceedings of the 2021 Conference of the North American Chapter of the Association for Computational Linguistics: Human Language Technologies, pages 4908–4915.

Updated

↑ back to top

This page in Markdown → https://md.research.elamur.ai/free/en/papers/zhou-2022-distributed-nli-learning-predict-human/