Elamur Research⌕
☰
home / papers / 2026 / What Exactly do Children Receive in Language Acquisition? A Case Study on CHILDES with Automated Detection of Filler-Gap Dependencies

What Exactly do Children Receive in Language Acquisition? A Case Study on CHILDES with Automated Detection of Filler-Gap Dependencies

Summary

Can children acquire filler-gap dependencies from the statistics of the speech they hear alone, or does this require innate grammatical knowledge? To test this, one needs to know what exactly children hear. Until now, such data have been annotated manually, which does not scale. The authors built an automatic detector for spoken English. It identifies three constructions: main wh-questions, embedded wh-questions, and relative clauses. For each, the detector determines where the element has been extracted from: the subject, object, or adjunct position. The method combines two kinds of syntactic parsing. Constituency parsing captures clause boundaries well and helps recognize the construction type. Dependency parsing helps locate the gap. The system was evaluated against human-annotated data and performs well in most categories. It was then applied to 57 English CHILDES corpora. This yields a description both of child-directed speech and of how children's own speech changes with age. The annotated data are suitable both for language acquisition research and for experiments with language models.

Key points

  • Tools: two parsers work in combination: spaCy's dependency parser and the Berkeley Neural Parser as implemented within spaCy. Constituency parsing proposes a hypothesis, and dependency relations (relcl, nsubj, nsubjpass) confirm it.
  • Annotation categories: 21 labels in total, for example SMQ, OMQ, ORC_reduced, and CC_ for questions with extraction across a clause boundary. Polar questions are also included in the annotation. Free relatives and infinitival relatives are deliberately excluded.
  • Data: 57 English CHILDES corpora. Construction frequencies and subject–object asymmetries were obtained for both adult and child speech. The code and dataset are published on GitHub (herbert-zhou/filler_gap_detector_childes).
  • Experiment: 15 LLaMA 360M models were trained on the BabyLM corpus with one of the constructions removed (filtered corpus training). Removing main wh-questions degraded performance on embedded questions and relative clauses. Removing either of the other two constructions caused no significant degradation on the rest. The authors conclude that more frequent constructions support generalization to rarer ones, but not the other way around.
  • Limitations: because of parser errors, the detector is imprecise, so it is suited to statistical tendencies rather than exact identification. Clefts and topicalization are not covered, and the system works only for English.

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

TopicsPsycholinguistics & language acquisition

Cite

Cite as Zhenghao Zhou, William Dai, Maya Viswanathan, Simon Charlow, R. Thomas McCoy and Robert Frank. 2026. What Exactly do Children Receive in Language Acquisition? A Case Study on CHILDES with Automated Detection of Filler-Gap Dependencies. Proceedings of the 30th Conference on Computational Natural Language Learning.

The paper itself, not this page. BibTeX from the library's record of the work. DOI 10.18653/v1/2026.conll-main.42 · arXiv 2603.02082

@inproceedings{zhou2026what,
  title = {{What Exactly do Children Receive in Language Acquisition? A Case Study on CHILDES with Automated Detection of Filler-Gap Dependencies}},
  author = {Zhenghao Zhou, William Dai, Maya Viswanathan, Simon Charlow, R. Thomas McCoy and Robert Frank},
  booktitle = {Proceedings of the 30th Conference on Computational Natural Language Learning},
  year = {2026},
  doi = {10.18653/v1/2026.conll-main.42},
  eprint = {2603.02082},
  archivePrefix = {arXiv},
  url = {https://doi.org/10.18653/v1/2026.conll-main.42}
}

References

39 references, 6 of them with a card of ours. ↗ leads to the work, «find» searches for one we have no id for yet.

  1. Ben Ambridge, Evan Kidd, Caroline F Rowland, and Anna L Theakston. 2015. The ubiquity of frequency effects in first language acquisition. Journal of Child Language, 42(2):239–273. find
  2. Emily Atkinson, Matthew W Wagers, Jeffrey Lidz, Colin Phillips, and Akira Omaki. 2018. Developing incrementality in filler-gap dependency processing. Cognition, 179:132–149. find
  3. Lois Bloom, Susan Merkin, and Janet Wootten. 1982. “wh”-Questions: Linguistic Factors That Contribute to the Sequence of Acquisition. Child development, pages 1084–1092. find
  4. Sasha Boguraev, Christopher Potts, and Kyle Mahowald. 2025. Causal Interventions Reveal Shared Structure Across English Filler–Gap Constructions. In Proceedings of the 2025 Conference on Empirical Methods in Natural Language Processing, pages 25032–25053. our card
  5. Susan R Braunwald. 1971. Mother-child communication: The function of maternal-language input. Word, 27(1-3):28–50. find
  6. Roger Brown. 1973. A first language: The early stages. Harvard University Press, Cambridge, MA. find
  7. Iona Carslaw, Sivan Milton, Nicolas Navarre, Ciyang Qing, and Wataru Uegaki. 2025. Automatic extraction of clausal embedding based on large-scale English text data. In Proceedings of the Society for Computation in Linguistics 2025, pages 322–332. find
  8. Chi-Yun Chang, Xueyang Huang, Humaira Nasir, Shane Storks, Olawale Akingbade, and Huteng Dai. 2025. Mind the Gap: How BabyLMs Learn Filler-Gap Dependencies. In Proceedings of the 2025 Conference on Empirical Methods in Natural Language Processing, pages 15060–15076. find
  9. Noam Chomsky. 1977. On Wh-movement. In Formal Syntax, pages 71–132. Academic Press, New York. find
  10. Noam Chomsky. 1986. Knowledge of Language: Its Nature, Origin, and Use. Convergence. Praeger, New York. find
  11. Atrey Desai and Sathvik Nair. 2026. Filling in the Mechanisms: How do LMs Learn Filler-Gap Dependencies under Developmental Constraints? Preprint, arXiv:2604.14459. find
  12. Holger Diessel and Michael Tomasello. 2005. A New Look at the Acquisition of Relative Clauses. Language, 81(1):1–25. find
  13. Michael C Frank. 2023. Bridging the data gap between children and large language models. Trends in Cognitive Sciences, 27(11):990–992. find
  14. Annie Gagliardi, Tara M Mease, and Jeffrey Lidz. 2016. Discontinuous development in the acquisition of filler-gap dependencies: Evidence from 15-and 20-month-olds. Language Acquisition, 23(3):234–260. find
  15. Roger Grosse, Juhan Bae, Cem Anil, Nelson Elhage, Alex Tamkin, Amirhossein Tajdini, Benoit Steiner, Dustin Li, Esin Durmus, Ethan Perez, Evan Hubinger, Kamilė Lukošiūtė, Karina Nguyen, Nicholas Joseph, Sam McCandlish, Jared Kaplan, and Samuel R. Bowman. 2023. Studying Large Language Model Generaliza… find
  16. Yiding Hao. 2020. Evaluating Attribution Methods using White-Box LSTMs. In Proceedings of the Third BlackboxNLP Workshop on Analyzing and Interpreting Neural Networks for NLP, pages 300–313, Online. Association for Computational Linguistics. find
  17. Matthew Honnibal, Ines Montani, Sofie Van Landeghem, and Adriane Boyd. 2020. spaCy: Industrial-strength Natural Language Processing in Python. find
  18. Abhinav Patil, Jaap Jumelet, Yu Ying Chiu, Andy La-pastora, Peter Shen, Lexie Wang, Clevis Willrich, and Shane Steinert-Threlkeld. 2024. Filtered corpus training (FiCT) shows that language models can generalize from indirect evidence. Transactions of the Association for Computational Linguistics, 12… find
  19. Katherine Howitt, Sathvik Nair, Allison Dods, and Robert Melvin Hopkins. 2024. Generalizations across filler-gap dependencies in neural language models. In Proceedings of the 28th conference on computational natural language learning, pages 269–279. find
  20. Lisa Pearl and Jon Sprouse. 2013. Computational models of acquisition for islands. Experimental Syntax and Islands Effects, pages 109–131. find
  21. Yaling Hsiao, Nicola J Dawson, Nilanjana Banerji, and Kate Nation. 2023. The nature and frequency of relative clauses in the language children hear and the language children read: A developmental cross-corpus analysis of English complex grammar. Journal of Child Language, 50(3):555–580. find
  22. Rodney D Huddleston, Geoffrey K Pullum, and Laurie Bauer. 2002. The Cambridge grammar of the English language. Cambridge University Press Cambridge. our card
  23. 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. find
  24. Nikita Kitaev, Steven Cao, and Dan Klein. 2019. Multilingual constituency parsing with self-attention and pre-training. In Proceedings of the 57th Annual Meeting of the Association for Computational Linguistics, pages 3499–3505, Florence, Italy. Association for Computational Linguistics. our card
  25. Nikita Kitaev and Dan Klein. 2018. Constituency parsing with a self-attentive encoder. In Proceedings of the 56th Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers), pages 2676–2686, Melbourne, Australia. Association for Computational Linguistics. our card
  26. Pang Wei Koh and Percy Liang. 2017. Understanding black-box predictions via influence functions. In International conference on machine learning, pages 1885–1894. PMLR. our card
  27. Nur Lan, Emmanuel Chemla, and Roni Katzir. 2024. Large language models and the argument from the poverty of the stimulus. Linguistic Inquiry, pages 1–28. find
  28. Julie Anne Legate and Charles D Yang. 2002. Empirical re-assessment of stimulus poverty arguments. The Linguistic Review, 19(1-2):151–162. find
  29. Geoffrey K Pullum and Barbara C Scholz. 2002. Empirical assessment of stimulus poverty arguments. The Linguistic Review, 19(1-2):9–50. find
  30. Alec Radford, Jeffrey Wu, Rewon Child, David Luan, Dario Amodei, and Ilya Sutskever. 2019. Language models are unsupervised multitask learners. OpenAI blog, 1(8):9. find
  31. Caroline F Rowland, Julian M Pine, Elena VM Lieven, and Anna L Theakston. 2003. Determinants of acquisition order in wh-questions: Re-evaluating the role of caregiver speech. Journal of Child Language, 30(3):609–635. find
  32. Alessandro Sanchez, Stephan C Meylan, Mika Braginsky, Kyle E MacDonald, Daniel Yurovsky, and Michael C Frank. 2019. childes-db: A flexible and reproducible interface to the child language data exchange system. Behavior Research Methods, 51(4):1928–1941. find
  33. Jon Sprouse, Ivano Caponigro, Ciro Greco, and Carlo Cecchetto. 2016. Experimental syntax and the variation of island effects in English and Italian. Natural Language & Linguistic Theory, 34(1):307–344. find
  34. Hugo Touvron, Thibaut Lavril, Gautier Izacard, Xavier Martinet, Marie-Anne Lachaux, Timothée Lacroix, Baptiste Rozière, Naman Goyal, Eric Hambro, Faisal Azhar, Aurelien Rodriguez, Armand Joulin, Edouard Grave, and Guillaume Lample. 2023. LLaMA: Open and Efficient Foundation Language Models. Preprint… our card
  35. Virginia Valian. 1991. Syntactic subjects in the early speech of American and Italian children. Cognition, 40(1-2):21–81. find
  36. Alex Warstadt, Aaron Mueller, Leshem Choshen, Ethan Wilcox, Chengxu Zhuang, Juan Ciro, Rafael Mos-quera, 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 th… find
  37. Brian MacWhinney. 2000. The CHILDES Project: Tools for Analyzing Talk. Volume II: The Database, 3rd edition. Lawrence Erlbaum Associates, Mahwah, NJ. find
  38. A Kanishka Misra and Kyle Mahowald. 2024. Language models learn rare phenomena from less rare phenomena: The case of the missing AANNs. In Proceedings of the 2024 Conference on Empirical Methods in Natural Language Processing, pages 913–929. find
  39. Section 3.2 illustrates an example of how constituency parsing complements dependency parsing An Example of How Dependency Parses Complement Constituency Parses find

↑ back to top

This page in Markdown → https://md.research.elamur.ai/free/en/papers/zhou-2026-what-exactly-do-children-receive/