Elamur Research⌕
☰
home / papers / 2023 / Language models are not naysayers: An analysis of language models on negation benchmarks

Language models are not naysayers: An analysis of language models on negation benchmarks

Paper · 2023 · Thinh Hung Truong, Timothy Baldwin, Karin Verspoor et al.

Summary

Earlier research showed that masked models such as BERT handle negation poorly. This paper asks whether that weakness persists in larger auto-regressive LLMs. The authors assembled six English benchmarks in which negation determines the correct answer and grouped them by the skill each one tests:

  • noticing negation in cloze-style factual queries
  • encoding negation in lexical relations (antonym vs. synonym)
  • drawing inferences under negation in NLI

They evaluated open-source GPT-Neo and OPT checkpoints (up to 6.7B parameters), GPT-3, InstructGPT and FLAN-T5-XXL. All evaluation was zero- or few-shot via prompting, with several prompt variants, including ones that explicitly point to negation.

Results:

  • Bigger base models were actually less sensitive to negation.
  • Antonym/synonym classification stayed near chance.
  • Most NLI scores fell below random.
  • Adding context or instructions usually hurt rather than helped.
  • Only instruction-tuned models, InstructGPT and FLAN-T5-XXL, showed real gains.

The authors conclude that current pre-training objectives do not capture negation. Better linguistic competence likely depends more on new training paradigms, such as instruction tuning, than on scale alone.

Key points

  • Benchmarks: MKR-NQ (a negated LAMA/ConceptNet cloze set), MWR, SAR, NegNLI, NaN-NLI and MoNLI. Code is published at github.com/joey234/llm-neg-bench.
  • Metrics: cloze tasks use Weighted Hit Rate (WHR), an error measure that counts confidence-weighted overlap with known wrong completions. Classification tasks use accuracy.
  • Inverse scaling on MKR-NQ: GPT-Neo-125M scored best, roughly on par with BERT-base. Prepending the affirmative sentence ("Contrasting" prompt) sharply increased errors.
  • Few-shot setup: prompts for SAR and MoNLI were prefixed with 10 random training examples.
  • Instruction tuning beats scale: FLAN-T5-XXL (11B) outperformed GPT-3 (175B) on most NLI tasks despite being about 16× smaller. InstructGPT still lagged on MoNLI, likely because it over-applied hypernymy reasoning.

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

🎓The University of MelbourneMelbourne · Australia100 works · 22,618 citations🎓Mohamed bin Zayed University of Artificial IntelligenceAbu Dhabi · United Arab Emirates124 works · 5,676 citations🎓RMIT UniversityMelbourne · Australia17 works · 703 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 2Research subjects and languages the work is about.Language models & pretraining · 3,320 worksEvaluation & benchmarks · 1,220 works

Cite

Cite as Thinh Hung Truong, Timothy Baldwin, Karin Verspoor and Trevor Cohn. 2023. Language models are not naysayers: An analysis of language models on negation benchmarks. arXiv:2306.08189.

The paper itself, not this page. BibTeX from the library's record of the work. DOI 10.48550/arxiv.2306.08189 · arXiv 2306.08189

@misc{truong2023language,
  title = {{Language models are not naysayers: An analysis of language models on negation benchmarks}},
  author = {Thinh Hung Truong, Timothy Baldwin, Karin Verspoor and Trevor Cohn},
  year = {2023},
  doi = {10.48550/arxiv.2306.08189},
  eprint = {2306.08189},
  archivePrefix = {arXiv},
  url = {https://doi.org/10.48550/arxiv.2306.08189}
}

References

49 references, 23 of them with a card of ours. ↗ leads to the work, «find» searches for one we have no id for yet.

  1. First, regarding the experimental settings, the WHR metrics used to evaluate cloze completion tasks are imperfect, as we discussed. Framing cloze completion tasks in the style of multiple-choice question answering to limit the options that models are evaluated on would be a good direction to follow… find
  2. Sid Black, Leo Gao, Phil Wang, Connor Leahy, and Stella Biderman. 2021. GPT-Neo: Large Scale Autoregressive Language Modeling with Mesh-Tensorflow. find
  3. 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 Computati…
  4. Tom Brown, Benjamin Mann, Nick Ryder, Melanie Subbiah, Jared D Kaplan, Prafulla Dhariwal, Arvind Neelakantan, Pranav Shyam, Girish Sastry, Amanda Askell, et al. 2020. Language models are few-shot learners. Advances in neural information processing systems, 33:1877–1901.
  5. Wendy W Chapman, Will Bridewell, Paul Hanbury, Gregory F Cooper, and Bruce G Buchanan. 2001. A simple algorithm for identifying negated findings and diseases in discharge summaries. Journal of Biomedical Informatics, 34(5):301–310. find
  6. Aakanksha Chowdhery, Sharan Narang, Jacob Devlin, Maarten Bosma, Gaurav Mishra, Adam Roberts, Paul Barham, Hyung Won Chung, Charles Sutton, Sebastian Gehrmann, et al. 2022. Palm: Scaling language modeling with pathways. arXiv preprint arXiv:2204.02311.
  7. Hyung Won Chung, Le Hou, Shayne Longpre, Barret Zoph, Yi Tay, William Fedus, Eric Li, Xuezhi Wang, Mostafa Dehghani, Siddhartha Brahma, et al. 2022. Scaling instruction-finetuned language models. arXiv preprint arXiv:2210.11416. find
  8. Ido Dagan, Oren Glickman, and Bernardo Magnini. 2005. The Pascal Recognising Textual Entailment challenge. In Machine learning challenges workshop, pages 177–190. Springer. find
  9. Allyson Ettinger. 2020. What BERT is not: Lessons from a new suite of psycholinguistic diagnostics for language models. Transactions of the Association for Computational Linguistics, 8:34–48.
  10. Leo Gao, Stella Biderman, Sid Black, Laurence Golding, Travis Hoppe, Charles Foster, Jason Phang, Horace He, Anish Thite, Noa Nabeshima, et al. 2020. The Pile: An 800gb dataset of diverse text for language modeling. arXiv preprint arXiv:2101.00027. find
  11. Atticus Geiger, Kyle Richardson, and Christopher Potts. 2020. Neural natural language inference models partially embed theories of lexical entailment and negation. In Proceedings of the Third BlackboxNLP Workshop on Analyzing and Interpreting Neural Networks for NLP, pages 163–173, Online. Associati…
  12. Hila Gonen, Srini Iyer, Terra Blevins, Noah A Smith, and Luke Zettlemoyer. 2022. Demystifying prompts in language models via perplexity estimation. arXiv preprint arXiv:2212.04037. find
  13. Md Mosharaf Hossain and Eduardo Blanco. 2022b. Leveraging affirmative interpretations from negation improves natural language understanding. In Proceedings of the 2022 Conference on Empirical Methods in Natural Language Processing, pages 5833– 5847, Abu Dhabi, United Arab Emirates. Association for C… find
  14. Md Mosharaf Hossain, Dhivya Chinnappa, and Eduardo Blanco. 2022. An analysis of negation in natural language understanding corpora. In Proceedings of the 60th Annual Meeting of the Association for Computational Linguistics (Volume 2: Short Papers), pages 716–723, Dublin, Ireland. Association for Com… find
  15. Md Mosharaf Hossain, Venelin Kovatchev, Pranoy Dutta, Tiffany Kao, Elizabeth Wei, and Eduardo Blanco. 2020. An analysis of natural language inference benchmarks through the lens of negation. In Proceedings of the 2020 Conference on Empirical Methods in Natural Language Processing (EMNLP), pages 9106…
  16. Arian Hosseini, Siva Reddy, Dzmitry Bahdanau, R Devon Hjelm, Alessandro Sordoni, and Aaron Courville. 2021. Understanding by understanding not: Modeling negation in language models. In Proceedings of the 2021 Conference of the North American Chapter of the Association for Computational Linguistics:…
  17. Joel Jang, Seonghyeon Ye, and Minjoon Seo. 2022a. Can large language models truly understand prompts? a case study with negated prompts. arXiv preprint arXiv:2209.12711. — [arXiv:2209.12711](https://arxiv.org/abs/2209.12711) find
  18. Myeongjun Jang and Thomas Lukasiewicz. 2023. Consistency analysis of chatgpt. arXiv preprint arXiv:2303.06273. find
  19. Reto Gubelmann and Siegfried Handschuh. 2022. Context matters: A pragmatic study of PLMs’ negation understanding. In Proceedings of the 60th Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers), pages 4602–4621, Dublin, Ireland. Association for Computational Lingui…
  20. Myeongjun Jang, Frank Mtumbuka, and Thomas Lukasiewicz. 2022b. Beyond distributional hypothesis: Let language models learn meaning-text correspondence. In Findings of the Association for Computational Linguistics: NAACL 2022, pages 2030– 2042, Seattle, United States. Association for Computational Li… find
  21. Ari Holtzman, Peter West, Vered Shwartz, Yejin Choi, and Luke Zettlemoyer. 2021. Surface form competition: Why the highest probability answer isn’t always right. In Proceedings of the 2021 Conference on Empirical Methods in Natural Language Processing, pages 7038–7051, Online and Punta Cana, Dominic…
  22. Ting Jiang, Jian Jiao, Shaohan Huang, Zihan Zhang, Deqing Wang, Fuzhen Zhuang, Furu Wei, Haizhen Huang, Denvy Deng, and Qi Zhang. 2022. Prompt-BERT: Improving BERT sentence embeddings with prompts. In Proceedings of the 2022 Conference on Empirical Methods in Natural Language Processing, pages 8826–… find
  23. Md Mosharaf Hossain and Eduardo Blanco. 2022a. Leveraging affirmative interpretations from negation improves natural language understanding. arXiv preprint arXiv:2210.14486. find
  24. Zhengbao Jiang, Frank F. Xu, Jun Araki, and Graham Neubig. 2020. How can we know what language models know? Transactions of the Association for Computational Linguistics, 8:423–438.
  25. Jaap Jumelet, Milica Denic, Jakub Szymanik, Dieuwke Hupkes, and Shane Steinert-Threlkeld. 2021. Language models use monotonicity to assess NPI licensing. In Findings of the Association for Computational Linguistics: ACL-IJCNLP 2021, pages 4958– 4969, Online. Association for Computational Linguistics… find
  26. Nora Kassner and Hinrich Schütze. 2020. Negated and misprimed probes for pretrained language models: Birds can talk, but cannot fly. In Proceedings of the 58th Annual Meeting of the Association for Computational Linguistics, pages 7811–7818, Online. Association for Computational Linguistics.
  27. Brian Lester, Rami Al-Rfou, and Noah Constant. 2021. The power of scale for parameter-efficient prompt tuning. In Proceedings of the 2021 Conference on Empirical Methods in Natural Language Processing, pages 3045–3059, Online and Punta Cana, Dominican Republic. Association for Computational Linguist…
  28. Vladislav Lialin, Kevin Zhao, Namrata Shivagunde, and Anna Rumshisky. 2022. Life after BERT: What do other muppets understand about language? In Proceedings of the 60th Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers), pages 3180–3193, Dublin, Ireland. Associat… find
  29. 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.
  30. Ian McKenzie, Alexander Lyzhov, Alicia Parrish, Ameya Prabhu, Aaron Mueller, Najoung Kim, Sam Bowman, and Ethan Perez. 2022. Inverse scaling prize: Round 1 winners. find
  31. George A Miller. 1998. WordNet: An electronic lexical database. MIT press. find
  32. Long Ouyang, Jeffrey Wu, Xu Jiang, Diogo Almeida, Carroll Wainwright, Pamela Mishkin, Chong Zhang, Sandhini Agarwal, Katarina Slama, Alex Gray, John Schulman, Jacob Hilton, Fraser Kelton, Luke Miller, Maddie Simens, Amanda Askell, Peter Welinder, Paul Christiano, Jan Leike, and Ryan Lowe. 2022. Trai… — Training language models to follow instructions with human feedback
  33. Fabio Petroni, Tim Rocktäschel, Sebastian Riedel, Patrick Lewis, Anton Bakhtin, Yuxiang Wu, and Alexander Miller. 2019. Language models as knowledge bases? In Proceedings of the 2019 Conference on Empirical Methods in Natural Language Processing and the 9th International Joint Conference on Natural…
  34. Alec Radford, Jeff Wu, Rewon Child, David Luan, Dario Amodei, and Ilya Sutskever. 2019. Language models are unsupervised multitask learners. OpenAI blog. find
  35. Abhilasha Ravichander, Matt Gardner, and Ana Marasović. 2022. CONDAQA: A contrastive reading comprehension dataset for reasoning about negation. arXiv preprint arXiv:2211.00295. find
  36. Joshua Robinson, Christopher Michael Rytting, and David Wingate. 2022. Leveraging large language models for multiple choice question answering. arXiv preprint arXiv:2210.12353.
  37. Robyn Speer, Joshua Chin, and Catherine Havasi. 2017. ConceptNet 5.5: An open multilingual graph of general knowledge. In Thirty-first AAAI conference on artificial intelligence. find
  38. Ieva Staliūnaitė and Ignacio Iacobacci. 2020. Compositional and lexical semantics in RoBERTa, BERT and DistilBERT: A case study on CoQA. In Proceedings of the 2020 Conference on Empirical Methods in Natural Language Processing (EMNLP), pages 7046–7056, Online. Association for Computational Linguis… find
  39. Thinh Truong, Timothy Baldwin, Trevor Cohn, and Karin Verspoor. 2022a. Improving negation detection with negation-focused pre-training. In Proceedings of the 2022 Conference of the North American Chapter of the Association for Computational Linguistics: Human Language Technologies, pages 4188–4193,… find
  40. Thinh Hung Truong, Yulia Otmakhova, Timothy Baldwin, Trevor Cohn, Jey Han Lau, and Karin Verspoor. 2022b. Not another negation benchmark: The NaN-NLI test suite for sub-clausal negation. In Proceedings of the 2nd Conference of the Asia-Pacific Chapter of the Association for Computational Linguistics…
  41. Alex Wang, Yada Pruksachatkun, Nikita Nangia, Amanpreet Singh, Julian Michael, Felix Hill, Omer Levy, and Samuel R. Bowman. 2019. SuperGLUE: A stickier benchmark for general-purpose language understanding systems. CoRR, abs/1905.00537.
  42. 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,…
  43. Hao Wang, Yangguang Li, Zhen Huang, Yong Dou, Lingpeng Kong, and Jing Shao. 2022. Sncse: Contrastive learning for unsupervised sentence embedding with soft negative samples. find
  44. Alex Warstadt, Yu Cao, Ioana Grosu, Wei Peng, Hagen Blix, Yining Nie, Anna Alsop, Shikha Bordia, Haokun Liu, Alicia Parrish, Sheng-Fu Wang, Jason Phang, Anhad Mohananey, Phu Mon Htut, Paloma Jeretic, and Samuel R. Bowman. 2019. Investigating BERT’s knowledge of language: Five analysis methods with N…
  45. Jason Wei, Yi Tay, Rishi Bommasani, Colin Raffel, Barret Zoph, Sebastian Borgeaud, Dani Yogatama, Maarten Bosma, Denny Zhou, Donald Metzler, et al. 2022a. Emergent abilities of large language models. Transactions on Machine Learning Research.
  46. Jason Wei, Yi Tay, and Quoc V Le. 2022b. Inverse scaling can become U-shaped. arXiv preprint arXiv:2211.02011. find
  47. Ethan Gotlieb Wilcox, Jon Gauthier, Jennifer Hu, Peng Qian, and Roger P. Levy. 2020. On the predictive power of neural language models for human real-time comprehension behavior. In Proceedings of the 42nd Annual Meeting of the Cognitive Science Society, page 1707–1713. find
  48. 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 Pa…
  49. Susan Zhang, Stephen Roller, Naman Goyal, Mikel Artetxe, Moya Chen, Shuohui Chen, Christopher Dewan, Mona Diab, Xian Li, Xi Victoria Lin, et al. 2022. OPT: Open pre-trained transformer language models. arXiv preprint arXiv:2205.01068.

Updated

↑ back to top

This page in Markdown → https://md.research.elamur.ai/free/en/papers/truong-2023-language-models-naysayers-analysis-language/