Elamur Research⌕
☰
home / papers / 2025 / Mechanisms vs. Outcomes: Probing for Syntax Fails to Explain Performance on Targeted Syntactic Evaluations

Mechanisms vs. Outcomes: Probing for Syntax Fails to Explain Performance on Targeted Syntactic Evaluations

Summary

The authors test whether the quality of syntactic probing can predict how a language model performs on targeted grammatical tests. They frame this as a contrast between "mechanisms" and "outcomes." A mechanism is what can be linearly extracted from hidden states. An outcome is the model's behavior on minimal pairs. The study covers 32 open transformers of up to 8B parameters: decoders, encoders, and encoder-decoders. For each of them, three linear probes that recover dependency trees were trained: structural, orthogonal structural, and head word. The probes were trained on the Penn Treebank with UD parses from Stanza. A control probe was also added, which reproduces distances between GloVe vectors and has nothing to do with syntax. Accuracy on BLiMP serves as the outcome measure. The relationship between probes and tests was examined in three ways: OLS regressions, an analysis of key tree edges, and a comparison of UUAS distributions. At no level of analysis did probe accuracy reliably predict model behavior. Therefore, a model's grammatical competence is better measured with external minimal-pair tests. This also supports building BLiMP-like resources for languages other than English.

Key points

  • The evaluation resource is BLiMP. It contains 67 paradigms across 13 linguistic phenomena, with 1000 pairs each. A model's answer counts as correct if the grammatical sentence receives the higher probability. For decoders, the sum of token log-probabilities is computed; for encoders and encoder-decoders, PLL. The probes were evaluated with the UUAS and UAS metrics.
  • Overall and per-phenomenon analysis. On the full set, none of the three syntactic probes yielded a statistically significant regression. Among individual phenomena, only anaphor agreement turned out to be significant. However, the non-syntactic control probe shows a similar relationship, so this result is most likely spurious.
  • Key tree edges. The probes were tested on recovering edges that matter for a specific pair, for example the subject–predicate relation. Successful recovery coincides with a correct answer on the pair in only 40–60% of cases, depending on the model.
  • UUAS distributions. In over 85% of paradigms, the UUAS distributions for correctly and incorrectly solved pairs largely overlap. Only in 26 of the 67 paradigms did a t-test find a significant difference for at least one model.
  • Limitations and code. The study covers only English and models of up to 8B. The authors propose transferring the methodology to MultiBLiMP 1.0 (over 128 thousand pairs, 101 languages) and TurBLiMP. Code: github.com/agananth/SyntaxMechanismsOutcomes.

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: Linguistics & languages

Cite

The paper itself, not this page. BibTeX from the library's record of the work. DOI 10.18653/v1/2025.emnlp-main.1712 · arXiv 2506.16678

@inproceedings{agarwal2025mechanisms,
  title = {{Mechanisms vs. Outcomes: Probing for Syntax Fails to Explain Performance on Targeted Syntactic Evaluations}},
  author = {Agarwal, Ananth and Jian, Jasper and Manning, Christopher D and Murty, Shikhar},
  booktitle = {Proceedings of the 2025 Conference on Empirical Methods in Natural Language Processing},
  year = {2025},
  doi = {10.18653/v1/2025.emnlp-main.1712},
  eprint = {2506.16678},
  archivePrefix = {arXiv},
  url = {https://doi.org/10.18653/v1/2025.emnlp-main.1712}
}

References

48 references, 14 of them with a card of ours. ↗ leads to the work, «find» searches for one we have no id for yet.

  1. Ebtesam Almazrouei, Hamza Alobeidli, Abdulaziz Al-shamsi, Alessandro Cappelli, Ruxandra Cojocaru, Mérouane Debbah, Étienne Goffinet, Daniel Hesslow, Julien Launay, Quentin Malartic, Daniele Mazzotta, Badreddine Noune, Baptiste Pannier, and Guilherme Penedo. 2023. The Falcon series of open language m… find our card
  2. Aryaman Arora, Dan Jurafsky, and Christopher Potts. 2024. CausalGym: Benchmarking causal interpretability methods on linguistic tasks. In Proceedings of the 62nd Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers), pages 14638–14663, Bangkok, Thailand. Association… find
  3. Nitin Bansal, Xiaohan Chen, and Zhangyang Wang. 2018. Can we gain more from orthogonality regularizations in training deep CNNs? Preprint, arXiv:1810.09102. — [arXiv:1810.09102](https://arxiv.org/abs/1810.09102) find
  4. Ezgi Başar, Francesca Padovani, Jaap Jumelet, and Arianna Bisazza. 2025. TurBLiMP: A Turkish benchmark of linguistic minimal pairs. Preprint, arXiv:2506.13487. find our card
  5. Yonatan Belinkov. 2022. Probing classifiers: Promises, shortcomings, and advances. Computational Linguistics, 48(1):207–219. find our card
  6. Stella Biderman, Hailey Schoelkopf, Quentin Gregory Anthony, Herbie Bradley, Kyle O’Brien, Eric Hal-lahan, Mohammad Aflah Khan, Shivanshu Purohit, USVSN Sai Prashanth, Edward Raff, and 1 others. 2023. Pythia: A suite for analyzing large language models across training and scaling. In International C… find
  7. Kevin S. Brown, Eiling Yee, Gitte Joergensen, Melissa Troyer, Elliot Saltzman, Jay Rueckl, James S. Magnu-son, and Ken McRae. 2023. Investigating the extent to which distributional semantic models capture a broad range of semantic relations. Cognitive Science, 47(5):e13291. find
  8. Kevin Clark, Urvashi Khandelwal, Omer Levy, and Christopher D. Manning. 2019. What does BERT look at? an analysis of BERT‘s attention. In Proceedings of the 2019 ACL Workshop BlackboxNLP: Analyzing and Interpreting Neural Networks for NLP, pages 276–286, Florence, Italy. Association for Computationa… find
  9. Kevin Clark, Minh-Thang Luong, Quoc V. Le, and Christopher D. Manning. 2020. Electra: Pre-training text encoders as discriminators rather than generators. In ICLR 2020. find
  10. 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
  11. Marie-Catherine de Marneffe, Christopher D. Manning, Joakim Nivre, and Daniel Zeman. 2021. Universal Dependencies. Computational Linguistics, 47(2):255–308. find our card
  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. Pablo Diego-Simón, Stéphane D’Ascoli, Emmanuel Chemla, Yair Lakretz, and Jean-Rémi King. 2024. A polar coordinate system represents syntax in large language models. Preprint, arXiv:2412.05571. — [arXiv:2412.05571](https://arxiv.org/abs/2412.05571) find
  14. Tiwalayo Eisape, Vineet Gangireddy, Roger Levy, and Yoon Kim. 2022. Probing for incremental parse states in autoregressive language models. In Findings of the Association for Computational Linguistics: EMNLP 2022, pages 2801–2813, Abu Dhabi, United Arab Emirates. Association for Computational Lingui… find
  15. Yanai Elazar, Shauli Ravfogel, Alon Jacovi, and Yoav Goldberg. 2021. Amnesic probing: Behavioral explanation with amnesic counterfactuals. Transactions of the Association for Computational Linguistics, 9:160– 175. find our card
  16. Atticus Geiger, Zhengxuan Wu, Hanson Lu, Josh Rozner, Elisa Kreiss, Thomas Icard, Noah Goodman, and Christopher Potts. 2022. Inducing causal structure for interpretable neural networks. In Proceedings of the 39th International Conference on Machine Learning, volume 162 of Proceedings of Machine Lear… find
  17. Aaron Grattafiori, Abhimanyu Dubey, Abhinav Jauhri, Abhinav Pandey, Abhishek Kadian, Ahmad Al-Dahle, Aiesha Letman, Akhil Mathur, Alan Schelten, Alex Vaughan, Amy Yang, Angela Fan, Anirudh Goyal, Anthony Hartshorn, Aobo Yang, Archi Mitra, Archie Sravankumar, Artem Korenev, Arthur Hinsvark, and 542 o… find
  18. Dirk Groeneveld, Iz Beltagy, Evan Walsh, Akshita Bhagia, Rodney Kinney, Oyvind Tafjord, Ananya Jha, Hamish Ivison, Ian Magnusson, Yizhong Wang, Shane Arora, David Atkinson, Russell Authur, Khy-athi Chandu, Arman Cohan, Jennifer Dumas, Yanai Elazar, Yuling Gu, Jack Hessel, and 24 others. 2024. OLMo:… find
  19. Suriya Gunasekar, Yi Zhang, Jyoti Aneja, Caio César Teodoro Mendes, Allie Del Giorno, Sivakanth Gopi, Mojan Javaheripi, Piero Kauffmann, Gustavo de Rosa, Olli Saarikivi, Adil Salim, Shital Shah, Harkirat Singh Behl, Xin Wang, Sébastien Bubeck, Ronen Eldan, Adam Tauman Kalai, Yin Tat Lee, and Yuanzhi… find
  20. Pengcheng He, Xiaodong Liu, Jianfeng Gao, and Weizhu Chen. 2021. Deberta: Decoding-enhanced bert with disentangled attention. In International Conference on Learning Representations. find our card
  21. John Hewitt and Percy Liang. 2019. Designing and interpreting probes with control tasks. In Proceedings of the 2019 Conference on Empirical Methods in Natural Language Processing and the 9th International Joint Conference on Natural Language Processing (EMNLP-IJCNLP), pages 2733–2743, Hong Kong, Chi… find
  22. 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
  23. Felix Hill, Roi Reichart, and Anna Korhonen. 2015. SimLex-999: Evaluating semantic models with (genuine) similarity estimation. Computational Linguistics, 41(4):665–695. find our card
  24. Jasper Jian and Siva Reddy. 2023. Syntactic substitutability as unsupervised dependency syntax. In Proceedings of the 2023 Conference on Empirical Methods in Natural Language Processing, pages 2341– 2360, Singapore. Association for Computational Linguistics. find
  25. Albert Q. Jiang, Alexandre Sablayrolles, Arthur Mensch, Chris Bamford, Devendra Singh Chaplot, Diego de las Casas, Florian Bressand, Gianna Lengyel, Guillaume Lample, Lucile Saulnier, Lélio Renard Lavaud, Marie-Anne Lachaux, Pierre Stock, Teven Le Scao, Thibaut Lavril, Thomas Wang, Timothée Lacroix,… find
  26. Jaap Jumelet, Leonie Weissweiler, Joakim Nivre, and Arianna Bisazza. 2025. Multiblimp 1.0: A massively multilingual benchmark of linguistic minimal pairs. Preprint, arXiv:2504.02768. find
  27. William A. Ladusaw. 1979. Polarity Sensitivity as Inherent Scope Relations. Doctoral dissertation, University of Texas, Austin. find
  28. Tomasz Limisiewicz and David Mareček. 2021. Introducing orthogonal constraint in structural probes. In Proceedings of the 59th Annual Meeting of the Association for Computational Linguistics and the 11th International Joint Conference on Natural Language Processing (Volume 1: Long Papers), pages 42… 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. CoRR, abs/1907.11692. find
  30. Mitchell P. Marcus, Beatrice Santorini, and Mary Ann Marcinkiewicz. 1993. Building a large annotated corpus of English: The Penn Treebank. Computational Linguistics, 19(2):313–330. find our card
  31. 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. Association for Computational Linguistics. find
  32. Max Müller-Eberstein, Rob van der Goot, and Barbara Plank. 2022. Probing for labeled dependency trees. In Proceedings of the 60th Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers), pages 7711–7726, Dublin, Ireland. Association for Computational Linguistics. find
  33. Shikhar Murty, Pratyusha Sharma, Jacob Andreas, and Christopher D. Manning. 2022. Characterizing intrinsic compositionality in transformers with tree projections. Preprint, arXiv:2211.01288. — [arXiv:2211.01288](https://arxiv.org/abs/2211.01288) find
  34. OpenAI, Aaron Hurst, Adam Lerer, Adam P. Goucher, Adam Perelman, Aditya Ramesh, Aidan Clark, AJ Ostrow, Akila Welihinda, Alan Hayes, Alec Radford, Aleksander M adry, Alex Baker-Whitcomb, Alex Beutel, Alex Borzunov, Alex Carney, Alex Chow, Alex Kirillov, Alex Nichol, and 400 others. 2024. Gpt-4o sys… find our card
  35. Jeffrey Pennington, Richard Socher, and Christopher Manning. 2014. GloVe: Global vectors for word representation. In Proceedings of the 2014 Conference on Empirical Methods in Natural Language Processing (EMNLP), pages 1532–1543, Doha, Qatar. Association for Computational Linguistics. find
  36. Peng Qi, Yuhao Zhang, Yuhui Zhang, Jason Bolton, and Christopher D. Manning. 2020. Stanza: A python natural language processing toolkit for many human languages. In Proceedings of the 58th Annual Meeting of the Association for Computational Linguistics: System Demonstrations, pages 101–108, Online.… find our card
  37. Open models based on gemini research and technology. Preprint, arXiv:2403.08295. Gemma Team, Morgane Riviere, Shreya Pathak, Pier Giuseppe Sessa, Cassidy Hardin, Surya Bhupatiraju, Léonard Hussenot, Thomas Mesnard, Bobak Shahriari, Alexandre Ramé, Johan Ferret, Peter Liu, Pouya Tafti, Abe Friesen, M… find
  38. Ian Tenney, Dipanjan Das, and Ellie Pavlick. 2019. BERT rediscovers the classical NLP pipeline. In Proceedings of the 57th Annual Meeting of the Association for Computational Linguistics, pages 4593– 4601, Florence, Italy. Association for Computational Linguistics. find our card
  39. Alec Radford, Jeff Wu, Rewon Child, David Luan, Dario Amodei, and Ilya Sutskever. 2019. Language models are unsupervised multitask learners. Technical report, OpenAI. find
  40. Hugo Touvron, Louis Martin, Kevin Stone, Peter Albert, Amjad Almahairi, Yasmine Babaei, Nikolay Bashlykov, Soumya Batra, Prajjwal Bhargava, Shruti Bhosale, Dan Bikel, Lukas Blecher, Cristian Canton Ferrer, Moya Chen, Guillem Cucurull, David Esiobu, Jude Fernandes, Jeremy Fu, Wenyin Fu, and 49 others… find
  41. Colin Raffel, Noam Shazeer, Adam Roberts, Katherine Lee, Sharan Narang, Michael Matena, Yanqi Zhou, Wei Li, and Peter J. Liu. 2020. Exploring the limits of transfer learning with a unified text-to-text transformer. Journal of Machine Learning Research, 21(140):1–67. find
  42. Mycal Tucker, Peng Qian, and Roger Levy. 2021. What if this modified that? Syntactic interventions with counterfactual embeddings. In Findings of the Association for Computational Linguistics: ACL-IJCNLP 2021, pages 862–875, Online. Association for Computational Linguistics. find
  43. Abhilasha Ravichander, Yonatan Belinkov, and Eduard Hovy. 2021. Probing the probing paradigm: Does probing accuracy entail task relevance? In Proceedings of the 16th Conference of the European Chapter of the Association for Computational Linguistics: Main Volume, pages 3363–3377, Online. Association… find
  44. Ben Wang and Aran Komatsuzaki. 2021. GPT-J-6B: A 6 Billion Parameter Autoregressive Language Model. https://github.com/ kingoflolz/mesh-transformer-jax. ↗
  45. Julian Salazar, Davis Liang, Toan Q. Nguyen, and Katrin Kirchhoff. 2020. Masked language model scoring. In Proceedings of the 58th Annual Meeting of the Association for Computational Linguistics. Association for Computational Linguistics. find our card
  46. 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. find
  47. Ekaterina Taktasheva, Maxim Bazhukov, Kirill Kon-cha, Alena Fenogenova, Ekaterina Artemova, and Vladislav Mikhailov. 2024. RuBLiMP: Russian benchmark of linguistic minimal pairs. In Proceedings of the 2024 Conference on Empirical Methods in Natural Language Processing, pages 9268–9299, Miami, Florid… find
  48. Gemma Team, Thomas Mesnard, Cassidy Hardin, Robert Dadashi, Surya Bhupatiraju, Shreya Pathak, Laurent Sifre, Morgane Rivière, Mihir Sanjay Kale, Juliette Love, Pouya Tafti, Léonard Hussenot, Pier Giuseppe Sessa, Aakanksha Chowdhery, Adam Roberts, Aditya Barua, Alex Botev, Alex Castro-Ros, Ambrose Sl… find

↑ back to top

This page in Markdown → https://md.research.elamur.ai/free/en/papers/agarwal-2025-mechanisms-vs-outcomes-probing-syntax/