Elamur Research⌕
☰
home / papers / 2024 / Language Models Learn Rare Phenomena from Less Rare Phenomena: The Case of the Missing AANNs

Language Models Learn Rare Phenomena from Less Rare Phenomena: The Case of the Missing AANNs

Paper · 2024 · Kanishka Misra and Kyle Mahowald

Summary

The paper asks whether language models trained on human-scale data handle rare grammar by memorizing it or by generalizing from related, more frequent patterns. The test case is the English AANN construction (e.g., "a beautiful five days"). The authors trained OPT-style transformers from scratch on the BabyLM-strict corpus (100M tokens) and on systematically edited versions of it. In some versions, AANN sentences were deleted. In others, they were replaced with scrambled variants. Further versions removed constructions hypothesized to support AANNs, such as singular treatment of plural measure phrases ("a few weeks is…"). Learning was measured with acceptability tests on novel AANN sentences, scored with SLOR.

Models that never saw a single AANN still preferred it over perturbed orders, well above chance. Removing the related constructions weakened this ability, and the drop exceeded what an equal loss of random data would cause. Models also generalized better when the AANNs they saw had more diverse fillers. The work shows that rare grammar can emerge from abstraction over commoner structures, which bears on debates about innate linguistic knowledge.

Key points

  • Setup: The models are OPT-architecture LMs with 12 layers and a 16,384-token vocabulary, trained up to 20 epochs with the transformers library. Each reported result is averaged over 3 random seeds.
  • Detection: A refined regex over POS-tagged BabyLM found 2,448 AANN instances, about 0.02% of the corpus. It caught 17 of 18 hand-annotated cases.
  • Counterfactual controls: Corpora with AANNs swapped for shuffled variants (e.g., ANAN, NAAN) did not yield comparable learning of those variants. This held even though AANN itself was learned without any exposure to it.
  • Related constructions: Holding out singular-agreeing measure noun phrases reduced the likelihood of unseen AANNs by 36.5% on average.
  • Slot variability and corrigendum: Greater variety in the adjective and noun slots improved generalization, while the numeral slot showed little difference. A corrigendum fixed a log-base bug in the KenLM 4-gram baseline, which makes that baseline stronger than first reported. It also corrected chance level to 20%, and the main conclusions remain unchanged.

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 Texas at AustinAustin · United States119 works · 57,950 citations

 Fields, topics and concepts

📚 Shelves 6The 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 worksPsychology · 1,649 worksDevelopmental and Educational Psychology · 523 works

🧭 Topics 1Research subjects and languages the work is about.Language models & pretraining · 3,894 works

Cite

Cite as Kanishka Misra and Kyle Mahowald. 2024. Language Models Learn Rare Phenomena from Less Rare Phenomena: The Case of the Missing AANNs. Proceedings of the 2024 Conference on Empirical Methods in Natural Language Processing.

The paper itself, not this page. BibTeX from the library's record of the work. DOI 10.18653/v1/2024.emnlp-main.53

@inproceedings{misra2024language,
  title = {{Language Models Learn Rare Phenomena from Less Rare Phenomena: The Case of the Missing AANNs}},
  author = {Kanishka Misra and Kyle Mahowald},
  booktitle = {Proceedings of the 2024 Conference on Empirical Methods in Natural Language Processing},
  year = {2024},
  doi = {10.18653/v1/2024.emnlp-main.53},
  url = {https://doi.org/10.18653/v1/2024.emnlp-main.53}
}

References

69 references, 27 of them with a card of ours. ↗ leads to the work, «find» searches for one we have no id for yet.

  1. Ahmed Abdelali, Francisco Guzman, Hassan Sajjad, and Stephan Vogel. 2014. The AMARA corpus: Building parallel language resources for the educational domain. In Proceedings of the Ninth International Conference on Language Resources and Evaluation (LREC’14), pages 1856–1862, Reykjavik, Iceland. Europ…
  2. R Harald Baayen. 2009. Corpus linguistics in morphology: Morphological productivity. Corpus Linguistics. An International Handbook, pages 900–919. find
  3. Marco Baroni. 2022. On the proper role of linguistically oriented deep net analysis in linguistic theorising. In Algebraic Structures in Natural Language, pages 1– 16. CRC Press. find
  4. Emily M Bender, Timnit Gebru, Angelina McMillan-Major, and Shmargaret Shmitchell. 2021. On the Dangers of Stochastic Parrots: Can Language Models Be Too Big? In Proceedings of the 2021 ACM Conference on Fairness, Accountability, and Transparency, pages 610–623. find
  5. Joan Bybee. 1995. Regular morphology and the lexicon. Language and Cognitive Processes, 10(5):425–455. find
  6. N. Chomsky. 1957. Syntactic Structures. The Hague: Mouton. find
  7. N. Chomsky. 1965. Aspects of the Theory of Syntax. MIT Press, Cambridge, MA. find
  8. N. Chomsky. 1986. Knowledge of language: Its nature, origin, and use. Praeger Publishers. find
  9. Noam Chomsky, Ian Roberts, and Jeffrey Watumull. 2023. Noam Chomsky: The False Promise of Chat-GPT. The New York Times. find
  10. Mary Dalrymple and Tracy Holloway King. 2019. An amazing four doctoral dissertations. Argumentum, 15(2019). Publisher: Debreceni Egyetemi Kiado. find
  11. 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
  12. Julie Kallini, Isabel Papadimitriou, Richard Futrell, Kyle Mahowald, and Christopher Potts. 2024. Mission: Impossible language models. arXiv:2401.06416.
  13. Ronen Eldan and Yuanzhi Li. 2023. TinyStories: How Small Can Language Models Be and Still Speak Coherent English? arXiv:2305.07759.
  14. Richard S Kayne. 2007. On the syntax of quantity in English. Linguistic theory and South Asian languages: Essays in honour of Ka Jayaseelan, 102:73. find
  15. Richard Futrell, Ethan Wilcox, Takashi Morita, Peng Qian, Miguel Ballesteros, and Roger Levy. 2019. Neural language models as psycholinguistic subjects: Representations of syntactic state. In Proceedings of the 2019 Conference of the North American Chapter of the Association for Computational Lingui…
  16. Caitlin Keenan. 2013. “A pleasant three days in Philadelphia”: Arguments for a pseudopartitive analysis. University of Pennsylvania Working Papers in Linguistics, 19(1):11. find
  17. Martin Gerlach and Francesc Font-Clos. 2020. A standardized Project Gutenberg corpus for statistical analysis of natural language and quantitative linguistics. Entropy, 22(1):126.
  18. Adele E Goldberg. 1995. Constructions: A Construction Grammar Approach to Argument Structure. University of Chicago Press. find
  19. Adele E Goldberg. 2005. Constructions at Work: The Nature of Generalization in Language. Oxford University Press. find
  20. Adele E Goldberg. 2019. Explain me this: Creativity, competition, and the partial productivity of constructions. Princeton University Press. find
  21. Kenneth Heafield. 2011. KenLM: Faster and smaller language model queries. In Proceedings of the Sixth Workshop on Statistical Machine Translation, pages 187–197, Edinburgh, Scotland. Association for Computational Linguistics.
  22. Felix Hill, Antoine Bordes, Sumit Chopra, and Jason Weston. 2016. The Goldilocks Principle: Reading Children’s Books with Explicit Memory Representations. In 4th International Conference on Learning Representations, ICLR 2016.
  23. Matthew Honnibal, Ines Montani, Sofie Van Lan-deghem, and Adriane Boyd. 2020. spaCy: Industrial-strength natural language processing in python. find
  24. Philip A. Huebner, Elior Sulem, Fisher Cynthia, and Dan Roth. 2021. BabyBERTa: Learning more grammar with small-scale child-directed language. In Proceedings of the 25th Conference on Computational Natural Language Learning, pages 624–646, Online. Association for Computational Linguistics.
  25. Najoung Kim, Tal Linzen, and Paul Smolensky. 2022. Uncontrolled Lexical Exposure Leads to Overestimation of Compositional Generalization in Pretrained Models. arXiv:2212.10769. — [arXiv:2212.10769](https://arxiv.org/abs/2212.10769) find
  26. Jey Han Lau, Alexander Clark, and Shalom Lappin. 2017. Grammaticality, acceptability, and probability: A probabilistic view of linguistic knowledge. Cognitive Science, 41(5):1202–1241. find
  27. Cara Su-Yi Leong and Tal Linzen. 2023. Language models can learn exceptions to syntactic rules. In Proceedings of the Society for Computation in Linguistics 2023, pages 133–144, Amherst, MA. Association for Computational Linguistics. find
  28. Cara Su-Yi Leong and Tal Linzen. 2024. Testing learning hypotheses using neural networks by manipulating learning data. arXiv:2407.04593.
  29. Bai Li, Zining Zhu, Guillaume Thomas, Frank Rudzicz, and Yang Xu. 2022. Neural reality of argument structure constructions. In Proceedings of the 60th Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers), pages 7410–7423, Dublin, Ireland. Association for Computatio… find
  30. Tal Linzen. 2020. How can we accelerate progress towards human-like linguistic generalization? In Proceedings of the 58th Annual Meeting of the Association for Computational Linguistics, pages 5210– 5217, Online. Association for Computational Linguistics.
  31. 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.
  32. Pierre Lison and Jörg Tiedemann. 2016. OpenSubtitles2016: Extracting large parallel corpora from movie and TV subtitles. In Proceedings of the Tenth International Conference on Language Resources and Evaluation (LREC’16), pages 923–929, Portoro!, Slovenia. European Language Resources Association (EL…
  33. 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:1907.11692.
  34. Kaiji Lu, Piotr Mardziel, Fangjing Wu, Preetam Amancharla, and Anupam Datta. 2020. Gender bias in neural natural language processing. Logic, Language, and Security: Essays Dedicated to Andre Scedrov on the Occasion of His 65th Birthday, pages 189–202.
  35. B. MacWhinney. 2000. The CHILDES project: Tools for analyzing talk. Lawrence Erlbaum Hillsdale, New Jersey. find
  36. Kyle Mahowald. 2023. A discerning several thousand judgments: GPT-3 rates the article + adjective + numeral + noun construction. In Proceedings of the 17th Conference of the European Chapter of the Association for Computational Linguistics, pages 265– 273, Dubrovnik, Croatia. Association for Computa… find
  37. Kyle Mahowald, Anna A Ivanova, Idan A Blank, Nancy Kanwisher, Joshua B Tenenbaum, and Evelina Fedorenko. 2024. Dissociating language and thought in large language models. Trends in Cognitive Sciences. find
  38. Rowan Hall Maudslay, Hila Gonen, Ryan Cotterell, and Simone Teufel. 2019. It’s all in the name: Mitigating gender bias with name-based counterfactual data substitution. In Proceedings of the 2019 Conference on Empirical Methods in Natural Language Processing and the 9th International Joint Conferenc…
  39. R. Thomas McCoy, Paul Smolensky, Tal Linzen, Jianfeng Gao, and Asli Celikyilmaz. 2023. How much do language models copy from their training data? evaluating linguistic novelty in text generation using RAVEN. Transactions of the Association for Computational Linguistics, 11:652–670.
  40. Kanishka Misra. 2022. minicons: Enabling flexible behavioral and representational analyses of transformer language models. arXiv:2203.13112.
  41. Kanishka Misra and Najoung Kim. 2023. Abstraction via exemplars? A representational case study on lexical category inference in BERT. In BUCLD 48: Proceedings of the 48th annual Boston University Conference on Language Development, Boston, USA. find
  42. Timothy J O’Donnell. 2015. Productivity and reuse in language: A theory of linguistic computation and storage. MIT Press. find
  43. Daniel N Osherson, Edward E Smith, Ormond Wilkie, Alejandro Lopez, and Eldar Shafir. 1990. Category-based Induction. Psychological Review, 97(2):185. find
  44. Abhinav Patil, Jaap Jumelet, Yu Ying Chiu, Andy Lapastora, Peter Shen, Lexie Wang, Clevis Willrich, and Shane Steinert-Threlkeld. 2024. Filtered Corpus Training (FiCT) Shows that Language Models can Generalize from Indirect Evidence. arXiv:2405.15750.
  45. Adam Pauls and Dan Klein. 2012. Large-scale syntactic language modeling with treelets. In Proceedings of the 50th Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers), pages 959–968, Jeju Island, Korea. Association for Computational Linguistics. find
  46. Lisa Pearl. 2022. Poverty of the stimulus without tears. Language Learning and Development, 18(4):415– 454. find
  47. Christopher Potts. 2023. Characterizing English Preposing in PP constructions. Ms., Stanford University. find
  48. Jakob Prange, Nathan Schneider, and Vivek Srikumar. 2021. Supertagging the long tail with tree-structured decoding of complex categories. Transactions of the Association for Computational Linguistics, 9:243– 260. find
  49. Geoffrey K Pullum. 2017. Theory, data, and the epistemology of syntax. In Grammatische Variation. Empirische Zugänge und theoretische Modellierung, pages 283–298. de Gruyter. find
  50. Geoffrey K Pullum and Barbara C Scholz. 2002. Empirical assessment of stimulus poverty arguments. The Linguistic Review, 19(1-2):9–50. find
  51. Alec Radford, Jeff Wu, Rewon Child, David Luan, Dario Amodei, and Ilya Sutskever. 2019. Language models are unsupervised multitask learners. OpenAI. find
  52. Roger Schwarzschild. 2011. Stubborn Distributivity, Multiparticipant Nouns and the Count/Mass Distinction. In Proceedings of NELS, volume 39, pages 661–678. Graduate Linguistics Students Association, University of Massachusetts. Issue: 2. find
  53. Koustuv Sinha, Robin Jia, Dieuwke Hupkes, Joelle Pineau, Adina Williams, and Douwe Kiela. 2021. Masked language modeling and the distributional hypothesis: Order word matters pre-training for little. In Proceedings of the 2021 Conference on Empirical Methods in Natural Language Processing, pages 288…
  54. Stephanie Solt. 2007. Two types of modified cardinals. In International Conference on Adjectives. Lille. find
  55. Andreas Stolcke, Klaus Ries, Noah Coccaro, Elizabeth Shriberg, Rebecca Bates, Daniel Jurafsky, Paul Taylor, Rachel Martin, Carol Van Ess-Dykema, and Marie Meteer. 2000. Dialogue act modeling for automatic tagging and recognition of conversational speech. Computational Linguistics, 26(3):339–374.
  56. Laura Suttle and Adele E Goldberg. 2011. The partial productivity of constructions as induction. Linguistics, 49(6):1237–1269. language models for the English comparative correlative. In Proceedings of the 2022 Conference on Empirical Methods in Natural Language Processing, pages 10859–10882, Abu Dh… find
  57. Harish Tayyar Madabushi, Laurence Romain, Dagmar Divjak, and Petar Milin. 2020. CxGBERT: BERT meets construction grammar. In Proceedings of the 28th International Conference on Computational Linguistics, pages 4020–4032, Barcelona, Spain (Online). International Committee on Computational Linguistics…
  58. Leonie Weissweiler, Abdullatif Köksal, and Hinrich Schütze. 2024. Hybrid human-LLM corpus construction and LLM evaluation for rare linguistic phenomena. arXiv:2403.06965.
  59. 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…
  60. Hugo Touvron, Louis Martin, Kevin Stone, Peter Albert, Amjad Almahairi, Yasmine Babaei, Nikolay Bashlykov, Soumya Batra, Prajjwal Bhargava, Shruti Bhosale, et al. 2023. Llama 2: Open foundation and fine-tuned chat models. arXiv:2307.09288.
  61. Yu-Hsiang Tseng, Cing-Fang Shih, Pin-Er Chen, Hsin-Yu Chou, Mao-Chang Ku, and Shu-Kai Hsieh. 2022. CxLM: A construction and context-aware language model. In Proceedings of the Thirteenth Language Resources and Evaluation Conference, pages 6361– 6369, Marseille, France. European Language Resources As… find
  62. Tim Veenboer and Jelke Bloem. 2023. Using collostructional analysis to evaluate BERT’s representation of linguistic constructions. In Findings of the Association for Computational Linguistics: ACL 2023, pages 12937–12951, Toronto, Canada. Association for Computational Linguistics. find
  63. Thomas Wolf, Lysandre Debut, Victor Sanh, Julien Chaumond, Clement Delangue, Anthony Moi, Pier-ric Cistac, Tim Rault, Remi Louf, Morgan Funtowicz, Joe Davison, Sam Shleifer, Patrick von Platen, Clara Ma, Yacine Jernite, Julien Plu, Canwen Xu, Teven Le Scao, Sylvain Gugger, Mariama Drame, Quentin Lho… — Transformers: State-of-the-Art Natural Language Processing
  64. Alex Warstadt. 2022. Artificial Neural Networks as Models of Human Language Acquisition. New York University. find
  65. Fei Xu and Joshua B Tenenbaum. 2007. Word learning as Bayesian inference. Psychological Review, 114(2):245. find
  66. Alex Warstadt, Aaron Mueller, Leshem Choshen, Ethan Wilcox, Chengxu Zhuang, Juan Ciro, Rafael Mosquera, Bhargavi Paranjabe, Adina Williams, Tal Linzen, and Ryan Cotterell. 2023. Findings of the BabyLM challenge: Sample-efficient pretraining on developmentally plausible corpora. In Proceedings of the… find
  67. Lucas Weber, Jaap Jumelet, Elia Bruni, and Dieuwke Hupkes. 2021. Language modelling as a multi-task problem. In Proceedings of the 16th Conference of the European Chapter of the Association for Computational Linguistics: Main Volume, pages 2049–2060, Online. Association for Computational Linguistics… find
  68. Jason Wei, Dan Garrette, Tal Linzen, and Ellie Pavlick. 2021. Frequency effects on syntactic rule learning in transformers. In Proceedings of the 2021 Conference on Empirical Methods in Natural Language Processing, pages 932–948, Online and Punta Cana, Dominican Republic. Association for Computation…
  69. 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:2205.01068.

Updated

↑ back to top

This page in Markdown → https://md.research.elamur.ai/free/en/papers/misra-2024-language-models-learn-rare-phenomena/