Elamur Research⌕
☰
home / papers / 2020 / Leap-Of-Thought: Teaching Pre-Trained Models to Systematically Reason Over Implicit Knowledge

Leap-Of-Thought: Teaching Pre-Trained Models to Systematically Reason Over Implicit Knowledge

Paper · 2020 · Alon Talmor, Oyvind Tafjord, Peter Clark et al.

Summary

Pre-trained language models hold a lot of world knowledge in their weights, but their reasoning over it is unreliable and hard to steer. Earlier work showed Transformers can reason over facts given explicitly in the input under a closed-world setting. This paper asks whether a model can be trained to combine explicitly provided statements with knowledge it already holds implicitly. The authors automatically build training data from true and false triples taken from ConceptNet, WordNet and Wikidata, turned into sentences with templates. Each example pairs a hypothesis with explicit statements, but key facts such as a hypernym link are left out, so the model must supply them from memory. Fine-tuning RoBERTa-Large on this data teaches taxonomic inference (hypernymy, meronymy) and counting. When the needed background knowledge is present in the LM, inference succeeds. Skills trained separately also combine in unseen multi-skill examples. This suggests users could fix model errors on the fly with a single natural-language statement, without retraining.

Key points

  • A motivating experiment fine-tuned RoBERTa on 50K TWENTY QUESTIONS (20Q) examples plus 100K RULETAKER examples. Without direct training for this skill, it reached 91.6% accuracy over 24 animal pairs that required mixing implicit size and taxonomy knowledge with given rules.
  • Inputs are formatted as [CLS] explicit knowledge [SEP] hypothesis [SEP] for a binary true/false decision. ESIM over GloVe serves as a non-contextual baseline.
  • On MULTI-SKILL-SET, the model trained on all four datasets scored 85.4 overall. Hypothesis-only inputs averaged about 40.2.
  • Skill composition generalized to unseen combinations:
    • The COUNTING+RULET.+20Q model reached 88.0 on hypernyms+counting.
    • The all-combined model scored 81.8 on hypernyms+sizes and 73.9 on hypernyms+age/year, even though these comparison skills were absent from training.
  • Code and data are released at github.com/alontalmor/LeapOfThought (NeurIPS 2020).

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

🔬Allen Institute for Artificial IntelligenceSeattle · United States304 works · 106,138 citations🎓Bar-Ilan UniversityRamat Gan · Israel146 works · 24,845 citations🎓Tel Aviv UniversityTel Aviv · Israel89 works · 11,326 citations

 Fields, topics and concepts

📚 Shelves 5The field of science this work belongs to (OpenAlex fields).Computer Science · 18,614 worksArtificial Intelligence · 17,965 worksArts and Humanities · 17,450 worksLanguage and Linguistics · 17,296 worksComputer Vision and Pattern Recognition · 3,486 works

Cite

Cite as Alon Talmor, Oyvind Tafjord, Peter Clark, Yoav Goldberg and Jonathan Berant. 2020. Leap-Of-Thought: Teaching Pre-Trained Models to Systematically Reason Over Implicit Knowledge. arXiv:2006.06609.

The paper itself, not this page. BibTeX from the library's record of the work. DOI 10.48550/arxiv.2006.06609 · arXiv 2006.06609

@misc{talmor2020leap,
  title = {{Leap-Of-Thought: Teaching Pre-Trained Models to Systematically Reason Over Implicit Knowledge}},
  author = {Alon Talmor, Oyvind Tafjord, Peter Clark, Yoav Goldberg and Jonathan Berant},
  year = {2020},
  doi = {10.48550/arxiv.2006.06609},
  eprint = {2006.06609},
  archivePrefix = {arXiv},
  url = {https://doi.org/10.48550/arxiv.2006.06609}
}

References

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

  1. Tom Mitchell, William Cohen, Estevam Hruschka, Partha Talukdar, Bishan Yang, Justin Betteridge, Andrew Carlson, Bhanava Dalvi, Matt Gardner, Bryan Kisiel, et al. Never-ending learning. Communications of the ACM, 61(5):103–115, 2018. find
  2. William Tunstall-Pedoe. True knowledge: Open-domain question answering using structured knowledge and inference. AI Magazine, 31(3):80–92, 2010. find
  3. Randall Davis. Interactive transfer of expertise: Acquisition of new inference rules. In IJCAI, 1977. find
  4. Peter Clark, Oyvind Tafjord, and Kyle Richardson. Transformers as soft reasoners over language. In IJCAI, 2020.
  5. A. Vaswani, N. Shazeer, N. Parmar, J. Uszkoreit, L. Jones, A. N. Gomez, L. Kaiser, and I. Polosukhin. Attention is all you need. arXiv preprint arXiv:1706.03762, 2017.
  6. M. E. Peters, M. Neumann, M. Iyyer, M. Gardner, C. Clark, K. Lee, and L. Zettlemoyer. Deep contextualized word representations. In North American Association for Computational Linguistics (NAACL), 2018.
  7. J. Devlin, M. Chang, K. Lee, and K. Toutanova. Bert: Pre-training of deep bidirectional transformers for language understanding. In North American Association for Computational Linguistics (NAACL), 2019.
  8. Yinhan Liu, Myle Ott, Naman Goyal, Jingfei Du, Mandar Joshi, Danqi Chen, Omer Levy, Mike Lewis, Luke Zettlemoyer, and Veselin Stoyanov. Roberta: A robustly optimized bert pretraining approach. arXiv preprint arXiv:1907.11692, 2019.
  9. A. Radford, K. Narasimhan, T. Salimans, and I. Sutskever. Improving language understanding by generative pre-training. Technical report, OpenAI, 2018. find
  10. Kyle Richardson and Ashish Sabharwal. What does my qa model know? devising controlled probes using expert knowledge. arXiv preprint arXiv:1912.13337, 2019.
  11. Alon Talmor, Yanai Elazar, Yoav Goldberg, and Jonathan Berant. olmpics–on what language model pre-training captures. arXiv preprint arXiv:1912.13283, 2019.
  12. Robert Speer, Joshua Chin, and Catherine Havasi. Conceptnet 5.5: An open multilingual graph of general knowledge. In AAAI, pages 4444–4451, 2017. find
  13. C. Fellbaum. WordNet: An Electronic Lexical Database. MIT Press, 1998. find
  14. D. Vrandečić and M. Krőtzsch. Wikidata: A free collaborative knowledgebase. Communications of the ACM, 57, 2014.
  15. Qian Chen, Xiaodan Zhu, Zhen-Hua Ling, Si Wei, Hui Jiang, and Diana Inkpen. Enhanced LSTM for natural language inference. In Proceedings of the 55th Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers), pages 1657–1668, Vancouver, Canada, 2017. Association for Comp… find
  16. J. Pennington, R. Socher, and C. D. Manning. GloVe: Global vectors for word representation. In Empirical Methods in Natural Language Processing (EMNLP), pages 1532–1543, 2014.
  17. Rowan Zellers, Yonatan Bisk, Roy Schwartz, and Yejin Choi. Swag: A large-scale adversarial dataset for grounded commonsense inference. In Proceedings of the 2018 Conference on Empirical Methods in Natural Language Processing (EMNLP), 2018. find
  18. Alon Talmor and Jonathan Berant. Multiqa: An empirical investigation of generalization and transfer in reading comprehension. 57th Annual Conference of the Association for Computational Linguistics (ACL), abs/1905.13453, 2019.
  19. Kyle Richardson, Hai Hu, Lawrence S Moss, and Ashish Sabharwal. Probing natural language inference models through semantic fragments. arXiv preprint arXiv:1909.07521, 2019.
  20. Ido Dagan, Dan Roth, Mark Sammons, and Fabio Massimo Zanzotto. Recognizing Textual Entailment: Models and Applications. Morgan and Claypool, 2013. find
  21. Samuel R. Bowman, Gabor Angeli, Christopher Potts, and Christopher D. Manning. A large annotated corpus for learning natural language inference. In EMNLP, 2015.
  22. Ankur P. Parikh, Oscar Täckström, Dipanjan Das, and Jakob Uszkoreit. A decomposable attention model for natural language inference. ArXiv, abs/1606.01933, 2016.
  23. Yi Tay, Anh Tuan Luu, and Siu Cheung Hui. Compare, compress and propagate: Enhancing neural architectures with alignment factorization for natural language inference. In EMNLP, 2018. find
  24. Suchin Gururangan, Swabha Swayamdipta, Omer Levy, Roy Schwartz, Samuel R. Bowman, and Noah A. Smith. Annotation artifacts in natural language inference data. In NAACL-HLT, 2018.
  25. Yixin Nie, Adina Williams, Emily Dinan, Mohit Bansal, Jason Weston, and Douwe Kiela. Adversarial nli: A new benchmark for natural language understanding. In ACL, 2020.
  26. Z. Yang, P. Qi, S. Zhang, Y. Bengio, W. W. Cohen, R. Salakhutdinov, and C. D. Manning. HotpotQA: A dataset for diverse, explainable multi-hop question answering. In Empirical Methods in Natural Language Processing (EMNLP), 2018.
  27. A. Talmor and J. Berant. The web as knowledge-base for answering complex questions. In North American Association for Computational Linguistics (NAACL), 2018. find
  28. J. Welbl, P. Stenetorp, and S. Riedel. Constructing datasets for multi-hop reading comprehension across documents. arXiv preprint arXiv:1710.06481, 2017.
  29. Akari Asai, Kazuma Hashimoto, Hannaneh Hajishirzi, Richard Socher, and Caiming Xiong. Learning to retrieve reasoning paths over wikipedia graph for question answering. ArXiv, abs/1911.10470, 2020.
  30. Yunxuan Xiao, Yanru Qu, Lin Qiu, Hao Zhou, Lei Li, Weinan Zhang, and Yong Yu. Dynamically fused graph network for multi-hop reasoning. In ACL, 2019.
  31. Sewon Min, Eric Wallace, Sameer Singh, Matt Gardner, Hannaneh Hajishirzi, and Luke Zettlemoyer. Compositional questions do not necessitate multi-hop reasoning. In ACL, 2019.
  32. Yuanpeng Li, Liang Zhao, JianYu Wang, and Joel Hestness. Compositional generalization for primitive substitutions. In EMNLP, 2019. find
  33. Guillaume Lample and Franccois Charton. Deep learning for symbolic mathematics. In ICLR, 2019.
  34. David Saxton, Edward Grefenstette, Felix Hill, and Pushmeet Kohli. Analysing mathematical reasoning abilities of neural models. In ICLR, 2019.
  35. Fabio Petroni, Patrick Lewis, Aleksandra Piktus, Tim Rocktäschel, Yuxiang Wu, Alexander H Miller, and Sebastian Riedel. How context affects language models’ factual predictions. arXiv preprint arXiv:2005.04611, 2020.
  36. Fabio Petroni, Tim Rocktäschel, Sebastian Riedel, Patrick Lewis, Anton Bakhtin, Yuxiang Wu, and Alexander Miller. 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 Langua…
  37. John W. McCarthy. Programs with common sense. In Proc. Tedding Conf. on the Mechanization of Thought Processes, pages 75–91, 1959. find
  38. Zhiting Hu, Xuezhe Ma, Zhengzhong Liu, Eduard H. Hovy, and Eric P. Xing. Harnessing deep neural networks with logic rules. In ACL, 2016.

Updated

↑ back to top

This page in Markdown → https://md.research.elamur.ai/free/en/papers/talmor-2020-leap-thought-teaching-pre-trained/