Elamur Research
☰
home / papers / 2020 / Information-Theoretic Probing for Linguistic Structure

Information-Theoretic Probing for Linguistic Structure

Written for the Research library (claude-opus-5-5) from the paper text. Original — aclanthology.org/2020.acl-main.420.pdf

Summary

The paper asks what probing classifiers actually measure when they are applied to neural representations. The authors recast probing as estimating the mutual information between a representation and a linguistic property. Under this view, a probe's cross-entropy gives a lower bound on that information, and a better probe tightens the bound. This contradicts the common advice to prefer simple probes. They also show, via the data-processing inequality, that a contextual embedder like BERT holds no more information about a property than the raw sentence does. This makes "ease of extraction" the more meaningful thing to formalize. As baselines, they introduce control functions, which are type-level representations that strip away context. They test multilingual BERT on POS tagging and dependency labeling across eleven typologically diverse languages using treebank data and deep MLP probes. BERT captures most of the uncertainty in both tasks, yet adds little beyond type-level baselines. This suggests that word-level POS probing reveals little about contextual syntax.

Key points

  • The eleven languages covered are Turkish, Korean, Tamil, Telugu, Urdu and Marathi, together with Finnish, Basque, Czech, Indonesian and English. Baselines are fastText and one-hot embeddings.
  • The probe is a ReLU MLP whose hidden size halves at each layer. For each estimate, 50 configurations are tried via random search, the one with the best validation cross-entropy is selected, and its test cross-entropy is used as the estimate.
  • POS: BERT beats the fastText control in only five of the eleven languages, and then by less than 9%. In some languages its gain is negative, which the authors attribute to estimation difficulty with finite training data.
  • Dependency labels: BERT outperforms type-level embeddings in every language. The largest gain is about 12% in English, and five languages show gains under 5%.
  • Code is released at github.com/rycolab/info-theoretic-probing.

Our summary of the paper, in our own words — not its text. The work belongs to its authors; read it at the links above.

Fields: Natural language processing

Cited by

133 papers of our library cite it (of the 140673 whose bibliographies we read); 4 of them have a page here, 129 not yet.

Cite

The paper itself, not this page. BibTeX from the library's record of the work. DOI 10.18653/v1/2020.acl-main.420

@inproceedings{pimentel2020information,
  title = {{Information-Theoretic Probing for Linguistic Structure}},
  author = {Pimentel, Tiago and Valvoda, Josef and Maudslay, Rowan Hall and Zmigrod, Ran and Williams, Adina and Cotterell, Ryan},
  booktitle = {Proceedings of the 58th Annual Meeting of the Association for Computational Linguistics},
  year = {2020},
  doi = {10.18653/v1/2020.acl-main.420},
  url = {https://doi.org/10.18653/v1/2020.acl-main.420}
}

References

30 references, 8 of them with a card of ours. ↗ leads to the work, «find» searches for one we have no id for yet.

  1. Guillaume Alain and Yoshua Bengio. 2017. Understanding intermediate layers using linear classifier probes. In 5th International Conference on Learning Representations. find our card
  2. Evan Archer, Il Memming Park, and Jonathan W. Pillow. 2014. Bayesian entropy estimation for countable discrete distributions. The Journal of Machine Learning Research, 15(1):2833–2868. find
  3. Yonatan Belinkov and James Glass. 2017. Analyzing hidden representations in end-to-end automatic speech recognition systems. In Advances in Neural Information Processing Systems, pages 2441–2451. find
  4. Yonatan Belinkov, Lluís Màrquez, Hassan Sajjad, Nadir Durrani, Fahim Dalvi, and James Glass. 2017. Evaluating layers of representation in neural machine translation on part-of-speech and semantic This is measured as the relative difference between H(T ) and H(T \| R). On average, this value is 88% an… find
  5. Piotr Bojanowski, Edouard Grave, Armand Joulin, and Tomas Mikolov. 2017. Enriching word vectors with subword information. TACL, 5:135–146. find
  6. 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… find our card
  7. Susan Carey. 1978. The child as word learner. In Linguistic Theory and Psychological Reality. MIT Press. find
  8. Coralie Chevallier, Ira A. Noveck, Tatjana Nazir, Lewis Bott, Valentina Lanzetti, and Dan Sperber. 2008. Making disjunctions exclusive. The Quarterly Journal of Experimental Psychology, 61(11):1741–1760. find
  9. Alexis Conneau, German Kruszewski, Guillaume Lample, Loïc Barrault, and Marco Baroni. 2018. What you can cram into a single $&!\#* vector: Probing sentence embeddings for linguistic properties. In Proceedings of the 56th Annual Meeting of the Association for Computational Linguistics (Volume 1: Long… find our card
  10. Thomas M. Cover and Joy A. Thomas. 2012. Elements of Information Theory. John Wiley & Sons. find
  11. George Cybenko. 1989. Approximation by superpositions of a sigmoidal function. Mathematics of Control, Signals and Systems, 2(4):303–314. find
  12. 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,… find
  13. Jerry A. Fodor, Thomas G. Bever, and Merrill F. Garrett. 1974. The Psychology of Language: An Introduction to Psycholinguistics and Generative Grammar. McGraw-Hill. find
  14. Victoria Fromkin, Robert Rodman, and Nina Hyams. 2018. An Introduction to Language. Cengage Learning. find
  15. Dan Garrette, Jason Mielens, and Jason Baldridge. 2013. Real-world semi-supervised learning of POS-taggers for low-resource languages. In Proceedings of the 51st Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers), pages 583–592, Sofia, Bulgaria. Association for C… find
  16. John Hewitt and Percy Liang. 2019. Designing and interpreting probes with control tasks. In Proceedings of the 2019 Conference on Empirical Methods Stephen C. Levinson. 2000. Presumptive Meanings: The Theory of Generalized Conversational Implicature. MIT Press. find
  17. Nelson F. Liu, Matt Gardner, Yonatan Belinkov, Matthew E. Peters, and Noah A. Smith. 2019. Linguistic knowledge and transferability of contextual representations. In Proceedings of the 2019 Conference of the North American Chapter of the Association for Computational Linguistics: Human Language Tech… find our card
  18. Aakanksha Naik, Abhilasha Ravichander, Norman Sadeh, Carolyn Rose, and Graham Neubig. 2018. Stress test evaluation for natural language inference. In Proceedings of the 27th International Conference on Computational Linguistics, pages 2340–2353, Santa Fe, New Mexico, USA. Association for Computation… find
  19. Sieb G. Nooteboom, Fred Weerman, and F. N. K. Wijnen. 2002. Storage and Computation in the Language Faculty. Springer Science & Business Media. find
  20. Matthew Peters, Mark Neumann, Mohit Iyyer, Matt Gardner, Christopher Clark, Kenton Lee, and Luke Zettlemoyer. 2018a. Deep contextualized word representations. In Proceedings of the 2018 Conference of the North American Chapter of the Association for Computational Linguistics: Human Language Technolo… find
  21. Matthew Peters, Mark Neumann, Luke Zettlemoyer, and Wen-tau Yih. 2018b. Dissecting contextual word embeddings: Architecture and representation. In Proceedings of the 2018 Conference on Empirical Methods in Natural Language Processing, pages 1499–1509, Brussels, Belgium. Association for Computational… find our card
  22. Slav Petrov, Dipanjan Das, and Ryan McDonald. 2012. A universal part-of-speech tagset. In Proceedings of the Eighth International Conference on Language Resources and Evaluation (LREC’12), pages 2089– 2096, Istanbul, Turkey. European Language Resources Association (ELRA). find
  23. Gözde Gül Sahin, Clara Vania, Ilia Kuznetsov, and Iryna Gurevych. 2019. LINSPECTOR: Multilingual probing tasks for word representations. CoRR, abs/1903.09442. find
  24. Naomi Saphra and Adam Lopez. 2019. Understanding learning dynamics of language models with SVCCA. 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 3257–3267, Minn… find
  25. 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. 2019. What do you learn from context? Probing for sentence structure in contextualized word representations. In 7th International Confere… find our card
  26. Elena Voita and Ivan Titov. 2020. Information-theoretic probing with minimum description length. arXiv preprint arXiv:2003.12298. find our card
  27. Thomas Wolf, Lysandre Debut, Victor Sanh, Julien Chaumond, Clement Delangue, Anthony Moi, Pier-ric Cistac, Tim Rault, Rémi Louf, Morgan Funtowicz, and Jamie Brew. 2019. Huggingface’s transformers: State-of-the-art natural language processing. CoRR, abs/1910.03771. find our card
  28. Yilun Xu, Shengjia Zhao, Jiaming Song, Russell Stewart, and Stefano Ermon. 2020. A theory of usable information under computational constraints. In International Conference on Learning Representations. find
  29. Daniel Zeman, Joakim Nivre, Mitchell Abrams, Noëmi Aepli, Željko Agić, Lars Ahrenberg, Gabrielė Alek-sandravičiūtė, Lene Antonsen, Katya Aplonova, Maria Jesus Aranzabe, Gashaw Arutie, Masayuki Asahara, Luma Ateyah, Mohammed Attia, Aitziber Atutxa, Liesbeth Augustinus, Elena Badmaeva, Miguel Bal… find
  30. Kelly Zhang and Samuel Bowman. 2018. Language modeling teaches you more than translation does: Lessons learned through auxiliary syntactic task analysis. In Proceedings of the 2018 EMNLP Workshop BlackboxNLP: Analyzing and Interpreting Neural Networks for NLP, pages 359–361, Brussels, Belgium. Assoc… find