Elamur Research⌕
☰
home / papers / 2023 / How Abstract Is Linguistic Generalization in Large Language Models? Experiments with Argument Structure

How Abstract Is Linguistic Generalization in Large Language Models? Experiments with Argument Structure

Paper · 2023 · Michael Wilson, Jackson Petty and Robert Frank

Summary

The paper asks whether pre-trained transformer LLMs encode how syntactic contexts relate to one another, beyond knowing which words fit a single context. The authors separate three levels: word-specific distributional knowledge (Type 0), verb-specific links between contexts that assign the same thematic role (Type 1), and verb-independent rules connecting such contexts, such as active object and passive subject (Type 2). They test BERT, DistilBERT and RoBERTa by adding unseen tokens to the vocabulary, fine-tuning each model on one construction only, and checking predictions in related constructions it never saw. Experiment 1 teaches novel nouns as arguments of spray/load verbs. Experiment 2 introduces a novel verb with new selectional preferences. The models handle Type 1 transfer well, mainly because the new noun embeddings move toward those of existing nouns with similar distributions. For the novel verb, generalization tracks linear order rather than structure. It succeeds when argument order is kept, as in polar questions, and fails when the order flips, as in passives. The authors argue this explains why LLMs need so much training data and why low-resource languages are at a disadvantage.

Key points

  • Models studied: BERT, DistilBERT and RoBERTa. Novel tokens were added whole, without subword splitting, and checked on every run.
  • Experiment 1 used the spray/load alternation with THAX as a novel theme and GORX as a novel goal. It had four fine-tuning sets (spray or load, crossed with theme-object or goal-object structure) of 12 sentences each.
  • Type 1 success came from a form of syntactic bootstrapping. The models assimilated the novel embeddings to embeddings of distributionally similar existing words.
  • Experiment 2 used larger fine-tuning sets of 288 sentences. Even so, performance dropped clearly whenever the test structure reversed the relative order of the verb's arguments.
  • Published in TACL vol. 11 (2023), pp. 1377–1395. Code and datasets are at github.com/clay-lab/structural-alternations.

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

🎓New York UniversityNew York · United States201 works · 100,180 citations🎓Yale UniversityNew Haven · United States95 works · 12,765 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.Language models & pretraining · 3,320 works

Cite

Cite as Michael Wilson, Jackson Petty and Robert Frank. 2023. How Abstract Is Linguistic Generalization in Large Language Models? Experiments with Argument Structure. Transactions of the Association for Computational Linguistics.

The paper itself, not this page. BibTeX from the library's record of the work. DOI 10.1162/tacl_a_00608

@article{wilson2023how,
  title = {{How Abstract Is Linguistic Generalization in Large Language Models? Experiments with Argument Structure}},
  author = {Michael Wilson, Jackson Petty and Robert Frank},
  journal = {Transactions of the Association for Computational Linguistics},
  year = {2023},
  doi = {10.1162/tacl\_a\_00608},
  url = {https://doi.org/10.1162/tacl_a_00608}
}

References

45 references, 15 of them with a card of ours. ↗ leads to the work, «find» searches for one we have no id for yet.

  1. Marc Brysbaert and Boris New. 2009. Moving beyond Kučera and Francis: A critical evaluation of current word frequency norms and the introduction of a new and improved word frequency measure for American English. Behavioral Research Methods, 41:977–990. https://doi.org/10.3758/BRM.41.4 .977, PubMed:… — [doi:10.3758/brm.41.4](https://doi.org/10.3758/brm.41.4) ↗
  2. Xingyu Cai, Jiaji Huang, Yuchen Bian, and Kenneth Church. 2021. Isotropy in the contextual embedding space: Clusters and manifolds. In Proceedings of the 9th International Conference on Learning Representations. find
  3. Shammur Absar Chowdhury and Roberto Zamparelli. 2019. An LSTM adaptation study of (un)grammaticality. In Proceedings of the 2019 ACL Workshop BlackboxNLP: Analyzing and Interpreting Neural Networks for NLP, pages 204–212, Florence, Italy. Association for Computational Linguistics. https://doi.org/1… find
  4. Jacob Devlin, Ming-Wei Chang, Kenton Lee, and Kristina Toutanova. 2019. BERT: Pre-training of deep bidirectional transformers for language understanding. In Proceedings of the 2019 Conference of the North American Chapter of the Association for Computational Linguistics: Human Language Technologies,…
  5. 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. https://doi.org/10 .1162/tacl_a_00298
  6. Robert M. French. 1999. Catastrophic forgetting in connectionist networks. Trends in Cognitive Sciences, 3(4):128–135. https://doi.org /10.1016/S1364-6613(99)01294-2, PubMed: 10322466 ↗
  7. Lila Gleitman. 1990. The structural sources of verb meaning. Language Acquisition, 1(1):3–55. https://doi.org/10.1207/s15327817la0101_2 ↗
  8. Lila Gleitman, Kimberly Cassidy, Rebecca Nappa, Anna Papafragou, and John Trueswell. 2005. Hard words. Language Learning and Development, 1(1):23–64. https://doi.org/10 .1207/s15473341lld0101_4 find
  9. Yoav Goldberg. 2019. Assessing BERT’s syntactic abilities. CoRR, abs/1901.05287v1.
  10. Jianping Gou, Baosheng Yu, Stephen J. Maybank, and Dacheng Tao. 2021. Knowledge distillation: A survey. International Journal of Computer Vision, 129(6):1789–1819. https://doi.org/10.1007/s11263-021-01453-z — [doi:10.1007/s11263-021-01453-z](https://doi.org/10.1007/s11263-021-01453-z) ↗
  11. Robert Hawkins, Minae Kwon, Dorsa Sadigh, and Noah Goodman. 2020a. Continual adaptation for efficient machine communication. In Proceedings of the 24th Conference on Computational Natural Language Learning, pages 408–419, Online. Association for Computational Linguistics. https://doi.org /10.18653/v… ↗
  12. Robert Hawkins, Takateru Yamakoshi, Thomas Griffiths, and Adele Goldberg. 2020b. Investigating representations of verb bias in neural language models. In Proceedings of the 2020 Conference on Empirical Methods in Natural Language Processing (EMNLP), pages 4653–4663, Online. Association for Computati… find
  13. John Hewitt and Christopher D. Manning. 2019. A structural probe for finding syntax in word representations. 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 4129…
  14. Ganesh Jawahar, Benoı̂t Sagot, and Djamé Seddah. 2019. What does BERT learn about the structure of language? In Proceedings of the 57th Annual Meeting of the Association for Computational Linguistics, pages 3651–3657, Florence, Italy. Association for Computational Linguistics. https://doi.org/10.18…
  15. Katharina Kann, Alex Warstadt, Adina Williams, and Samuel R. Bowman. 2019. Verb argument structure alternations in word and sentence embeddings. In Proceedings of the Society for Computation in Linguistics (SCiL) 2019, pages 287–297. https://doi.org/10.7275/q5js-4y86 ↗
  16. Tal Linzen, Emmanuel Dupoux, and Yoav Goldberg. 2016. Assessing the ability of LSTMs to learn syntax-sensitive dependencies. Transactions of the Association for Computational Linguistics, 4:521–535. https://doi.org/10.1162/tacl_a_00115 ↗
  17. Najoung Kim and Paul Smolensky. 2021. Testing for grammatical category abstraction in neural language models. In Proceedings of the Society for Computation in Linguistics (SCiL) 2021, pages 467–470, Online. Association for Computational Linguistics. find
  18. 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. CoRR, abs/1907.11692v1.
  19. Karim Lasri, Alessandro Lenci, and Thierry Poibeau. 2022. Does BERT really agree? Fine-grained analysis of lexical dependence on a syntactic task. In Findings of the Association for Computational Linguistics: ACL 2022, pages 2309–2315, Dublin, Ireland. Association for Computational Linguistics. http… find
  20. Sharid Loáiciga, Luca Bevacqua, and Christian Hardmeier. 2021. Unsupervised discovery of unaccusative and unergative verbs. CoRR, abs/2111.00808v1. — [arXiv:2111.00808](https://arxiv.org/abs/2111.00808) find
  21. Quentin Lhoest, Albert Villanova del Moral, Yacine Jernite, Abhishek Thakur, Patrick von Platen, Suraj Patil, Julien Chaumond, Mariama Drame, Julien Plu, Lewis Tunstall, Joe Davison, Mario Šaško, Gunjan Chhablani, Bhavitvya Malik, Simon Brandeis, Teven Le Scao, Victor Sanh, Canwen Xu, Nicolas Patr… — Datasets: A Community Library for Natural Language Processing
  22. Yongjie Lin, Yi Chern Tan, and Robert Frank. 2019. Open sesame: Getting inside BERT’s Rebecca Marvin and Tal Linzen. 2018. Targeted syntactic evaluation of language models. In Proceedings of the 2018 Conference on Empirical Methods in Natural Language Processing, pages 1192–1202, Brussels, Belgium.… find
  23. Michael McCloskey and Neal J. Cohen. 1989. Catastrophic interference in connectionist networks: The sequential learning problem. In Psychology of Learning and Motivation, volume 24, pages 109–165. Elsevier. https://doi.org /10.1016/S0079-7421(08)60536-8 ↗
  24. Eleni Metheniti, Tim Van de Cruys, and Nabil Hathout. 2020. How relevant are selectional preferences for transformer-based language models? In Proceedings of the 28th International Conference on Computational Linguistics, pages 1266–1278, Barcelona, Spain (Online). International Committee on Computa… find
  25. Marius Mosbach, Maksym Andriushchenko, and Dietrich Klakow. 2021. On the stability of fine-tuning BERT: Misconceptions, explanations, and strong baselines. In Proceedings of the 9th International Conference on Learning Representations. find
  26. Aaron Mueller, Robert Frank, Tal Linzen, Luheng Wang, and Sebastian Schuster. 2022. Coloring the blank slate: Pre-training imparts a hierarchical inductive bias to sequence-to-sequence models. In Findings of the Association for Computational Linguistics: ACL 2022, pages 1352–1368, Dublin, Ireland. A… find
  27. Benjamin Newman, Kai-Siang Ang, Julia Gong, and John Hewitt. 2021. Refining targeted syntactic evaluation of language models. In Proceedings of the 2021 Conference of the North American Chapter of the Association for Computational Linguistics: Human Language Technologies, pages 3710–3723, Online. As… find
  28. Ellie Pavlick. 2022. Semantic structure in deep learning. Annual Review of Linguistics, 8(1):447–471. https://doi.org/10.1146/annurev-linguistics-031120-122924 ↗
  29. Jackson Petty and Robert Frank. 2021. Transformers generalize linearly. CoRR, abs/2109 .12036v1. find
  30. Jackson Petty, Michael Wilson, and Robert Frank. 2022. Do language models learn position-role mappings? In Proceedings of the 46th Annual Boston University Conference on Language Development, volume 2, pages 657–671, Somerville, MA. Cascadilla Press. find
  31. Steven Pinker. 1989. Learnability and Cognition: The Acquisition of Argument Structure, 2013 new edition, volume 27 of Learning, Development, and Conceptual Change. MIT Press, Cambridge, MA. find
  32. Victor Sanh, Lysandre Debut, Julien Chaumond, and Thomas Wolf. 2020. DistilBERT, a distilled version of BERT: Smaller, faster, cheaper and lighter. CoRR, abs/1910.01108v4. find
  33. Ryohei Sasano and Anna Korhonen. 2020. Investigating word-class distributions in word vector spaces. In Proceedings of the 58th Annual Meeting of the Association for Computational Linguistics, pages 3657–3666, Online. find
  34. Association for Computational Linguistics. https://doi.org/10.18653/v1/2020 .acl-main.337 — [doi:10.18653/v1/2020](https://doi.org/10.18653/v1/2020) ↗
  35. Thibault Sellam, Steve Yadlowsky, Ian Tenney, Jason Wei, Naomi Saphra, Alexander D’Amour, Tal Linzen, Jasmijn Bastings, Iulia Raluca Turc, Jacob Eisenstein, Dipanjan Das, and Ellie Pavlick. 2022. The MultiBERTs: BERT reproductions for robustness analysis. In Proceedings of the 10th International Con… find
  36. Ian Tenney, Dipanjan Das, and Ellie Pavlick. 2019a. BERT rediscovers the classical NLP pipeline. In Proceedings of the 57th Annual Meeting of the Association for Computational Linguistics. https://doi.org/10 .18653/v1/P19-1452
  37. Ian Tenney, Patrick Xia, Berlin Chen, Alex Wang, Adam Poliak, R. Thomas McCoy, Najoung Kim, Benjamin Van Durme, Samuel R. Bowman, Dipanjan Das, and Ellie Pavlick. 2019b. What do you learn from context? Probing for sentence structure in contextualized word representations. In Proceedings of the 7th I…
  38. Tristan Thrush, Ethan Wilcox, and Roger Levy. 2020. Investigating novel verb learning in BERT: Selectional preference classes and alternation-based syntactic generalization. In Proceedings of the Third BlackboxNLP Workshop on Analyzing and Interpreting Neural Networks for NLP, pages 265–275, Online.… find
  39. William Timkey and Marten van Schijndel. 2021. All bark and no bite: Rogue dimensions in transformer language models obscure representational quality. In Proceedings of the 2021 Conference on Empirical Methods in Natural Language Processing, pages 4527–4546, Online and Punta Cana, Dominican Republic…
  40. Alex Warstadt and Samuel R. Bowman. 2020. Linguistic analysis of pretrained sentence encoders with acceptability judgments. CoRR, abs/1901.03438v4. find
  41. Alex Warstadt, Alicia Parrish, Haokun Liu, Anhad Mohananey, Wei Peng, Sheng-Fu Wang, and Samuel R. Bowman. 2020. BLiMP: The Benchmark of Linguistic Minimal Pairs for English. Transactions of the Association for Computational Linguistics, 8:377–392. https://doi.org/10.1162/tacl a 00321 ↗
  42. Ethan Wilcox, Roger Levy, Takashi Morita, and Richard Futrell. 2018. What do RNN language models learn about filler–gap dependencies? In Proceedings of the 2018 EMNLP Workshop BlackboxNLP: Analyzing and Interpreting Neural Networks for NLP, pages 211–221, Brussels, Belgium. Association for Computati…
  43. Thomas Wolf, Lysandre Debut, Victor Sanh, Julien Chaumond, Clement Delangue, Anthony Moi, Pierric Cistac, Tim Rault, Rémi Louf, Morgan Funtowicz, and Jamie Brew. 2020. Huggingface’s transformers: State-of-the-art natural language processing (version 4.12.5). CoRR, abs/1910.03771v5.
  44. Ryo Yoshida and Yohei Oseki. 2022. Learning argument structures with recurrent neural network grammars. In Proceedings of the Society for Computation in Linguistics (SCiL) 2022, volume 5, pages 101–111. find
  45. Xunjie Zhu and Gerard de Melo. 2020. Sentence analogies: Linguistic regularities in sentence embeddings. In Proceedings of the 28th International Conference on Computational Linguistics, pages 3389–3400, Barcelona, Spain (Online). International Committee on Computational Linguistics. https://doi.org… ↗

Updated

↑ back to top

This page in Markdown → https://md.research.elamur.ai/free/en/papers/wilson-2023-how-abstract-is-linguistic-generalization/