Elamur Research⌕
☰
home / papers / 2026 / Causal Inferences Are Driven by Noun Concept Specificity Evidence from Self-Paced Reading and Large Language Model Surprisal

Causal Inferences Are Driven by Noun Concept Specificity Evidence from Self-Paced Reading and Large Language Model Surprisal

Summary

Psycholinguists debate whether readers infer causal links between clauses only after completing syntactic and semantic analysis (a modular view) or while they are still reading (an interactive view). Earlier work found that when a relative clause can be read as the reason for an event, a following because-clause becomes harder to integrate, because the slot for an explanation is already filled. Schlotterbeck and Barth ask whether the specificity of the nouns involved also shapes this process.

They ran a German self-paced reading experiment that varied how specific the noun phrases were. The test sentences combined an implicit causality verb, a relative clause open to a causal reading, and an explicit because-clause. They also computed GPT-2 surprisal for these items, for the English materials of Hoek et al. (2021), and for a German translation of those materials.

Noun specificity reliably changed reading times inside the because-clause. The authors explain this with frame semantics: specific nouns evoke narrower scenarios, which produce stronger causal expectations. GPT-2 surprisal reproduced the key interaction but diverged from human reading times in places. This matters for judging how well surprisal captures discourse-level processing.

Key points

  • The study extends Rohde et al. (2011), a self-paced reading study of relative-clause attachment with implicit causality (IC) verbs. It also extends Hoek et al. (2021), an eye-tracking study crossing causal vs. neutral relative clauses with because vs. "and so" connectors, which was interpreted through the Empty Slot Theory.
  • In the new German self-paced reading study, NP specificity affected processing of the explicit because-clause. This supports the idea that causal expectations at the discourse level are built incrementally.
  • Some effects appeared relatively late in the because-clause rather than on the connective itself. The authors therefore favour a nuanced interactive model over a strictly modular one.
  • GPT-2 was chosen because medium-sized models have been reported to fit reading times better than larger ones (Oh and Schuler, 2023). Its surprisal captured the critical interaction in the German items, the Hoek et al. items, and their translation, but with substantial deviations from human data. Testing internal-layer surprisal (Kuribayashi et al., 2025) is proposed as future work.
  • The paper builds on Raphael Barth's 2025 M.A. thesis (University of Tübingen). It was funded by the DFG (grant 538184825) and published at CMCL @ LREC 2026, pages 276–285.

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 10.63317/27uaw5cm26ib

@inproceedings{schlotterbeck2026causal,
  title = {{Causal Inferences Are Driven by Noun Concept Specificity Evidence from Self-Paced Reading and Large Language Model Surprisal}},
  author = {Schlotterbeck, Fabian and Barth, Raphael},
  booktitle = {Proceedings of the 15th Workshop on Cognitive Modeling and Computational Linguistics},
  year = {2026},
  doi = {10.63317/27uaw5cm26ib},
  url = {https://doi.org/10.63317/27uaw5cm26ib}
}

References

23 references, 3 of them with a card of ours. ↗ leads to the work, «find» searches for one we have no id for yet.

  1. Charles Fillmore. 1982. Frame semantics. Linguistics in the Morning Calm. Seoul, pages 111–137. find
  2. Alex Fine, Ting Qian, T. Florian Jaeger, and Robert Jacobs. 2010. Syntactic adaptation in language comprehension. In Proceedings of the 2010 Workshop on Cognitive Modeling and Computational Linguistics, pages 18–26, Uppsala, Sweden. Association for Computational Linguistics. find
  3. Jerry A Fodor. 1983. The modularity of mind. MIT press, Cambridge, MA. find
  4. Lyn Frazier. 1987. Sentence processing: A tutorial review. In Max Coltheart, editor, Attention and Performance 12: The Psychology of Reading, pages 559–586. Lawrence Erlbaum Associates. find
  5. John Hale. 2001. A probabilistic earley parser as a psycholinguistic model. In Second Meeting of the North American chapter of the Association for Computational Linguistics. find our card
  6. Jet Hoek, Hannah Rohde, Jacqueline Evers-Vermeul, and Ted J.M. Sanders. 2021. Expectations from relative clauses: Real-time coherence updates in discourse processing. Cognition, 210:104581. find
  7. Marcel Adam Just and Patricia A Carpenter. 1976. Eye fixations and cognitive processes. Cognitive Psychology, 8(4):441–480. find
  8. Suhas Arehalli, Brian Dillon, and Tal Linzen. 2022. Syntactic surprisal from neural models predicts, but underestimates, human processing difficulty from syntactic ambiguities. In Proceedings of the 26th Conference on Computational Natural Language Learning (CoNLL 2022). Association for Computationa… find
  9. Raphael Barth. 2025. Psycholinguistische Untersuchung des Einflusses von Nomi-nalphrasenkonzepten auf die kausale Inferenz-bildung. Master’s thesis, University of Tübingen, Tübingen, Germany. find
  10. Douglas Bates, Martin Mächler, Ben Bolker, and Steve Walker. 2015. Fitting linear mixed-effects models using lme4. Journal of Statistical Software, 67(1):1–48. find our card
  11. Oliver Bott and Torgrim Solstad. 2023. The production of referring expressions is influenced by the likelihood of next mention. Quarterly Journal of Experimental Psychology, 76(10):2256–2284. find
  12. Noam Chomsky. 1965. Aspects of the Theory of Syntax. The MIT Press, Cambridge. find
  13. Jonathan Cohen and Andrew Kehler. 2021. Conversational eliciture. Philosophers Imprint, 21(12):1– 26. find
  14. Tatsuki Kuribayashi, Yohei Oseki, Souhaib Ben Taieb, Kentaro Inui, and Timothy Baldwin. 2025. Large language models are human-like internally. Transactions of the Association for Computational Linguistics, 13:1743–1766. find
  15. Roger Levy. 2008. Expectation-based syntactic comprehension. Cognition, 106(3):1126–1177. find
  16. Ken McRae, Michael J Spivey-Knowlton, and Michael K Tanenhaus. 1998. Modeling the influence of thematic fit (and other constraints) in online sentence comprehension. Journal of Memory and Language, 38(3):283–312. find
  17. Byung-Doh Oh and William Schuler. 2022. Entropy-and distance-based predictors from GPT-2 attention patterns predict reading times over and above GPT-2 surprisal. In Proceedings of the 2022 Conference on Empirical Methods in Natural Language Processing, pages 9324–9334, Abu Dhabi, United Arab Emirate… find
  18. Byung-Doh Oh and William Schuler. 2023. Why does surprisal from larger transformer-based language models provide a poorer fit to human reading times? Transactions of the Association for Computational Linguistics, 11:336–350. find our card
  19. Hannah Rohde, Roger Levy, and Andrew Kehler. 2011. Anticipating explanations in relative clause processing. Cognition, 118(3):339–358. find
  20. Cory Shain, Clara Meister, Tiago Pimentel, Ryan Cotterell, and Roger Levy. 2024. Large-scale evidence for logarithmic effects of word predictability on reading time. Proceedings of the National Academy of Sciences, 121(10):e2307876121. find
  21. Nathaniel J. Smith and R. Levy. 2013. The effect of word predictability on reading time is logarithmic. Cognition, 128:302–19. find
  22. Lorraine K. Tyler and William D. Marslen-Wilson. 1977. The on-line effects of semantic context on syntactic processing. Journal of Verbal Learning and Verbal Behavior, 16(6):683–692. find
  23. Ashish Vaswani, Noam Shazeer, Niki Parmar, Jakob Uszkoreit, Llion Jones, Aidan N Gomez, Łukasz Kaiser, and Illia Polosukhin. 2017. Attention is all you need. In Advances in Neural Information Processing Systems, pages 5998– 6008. find

↑ back to top

This page in Markdown → https://md.research.elamur.ai/free/en/papers/schlotterbeck-2026-causal-inferences-are-driven-noun/