Elamur Research⌕
☰
home / papers / 2025 / Performance and competence intertwined: A computational model of the Null Subject stage in English-speaking children

Performance and competence intertwined: A computational model of the Null Subject stage in English-speaking children

Paper · 2025 · Soumik Dey and William Sakas

Summary

The paper asks why English-learning children go through a phase, lasting until roughly age four, in which they often drop subjects. English requires subjects, so a learner should not drift toward the broader null-subject grammar. Orfitelli and Hyams (2012) proposed a performance-based explanation: toddlers often take subjectless imperatives as declaratives, and this misreading temporarily supports a null-subject grammar.

Dey and Sakas test that idea computationally. They define a developmental "performance parameter", the Illocution Ambiguity Resolution Coefficient (IARC), which is the probability that a child reads a subjectless imperative correctly. They fit IARC to O&H's age-group data. They then feed input from the CoLAG English-like domain to a modified Variational Learner, the Superset/Subset Variational Learner (SSVL). Unlike the original VL, SSVL avoids settling on a superset grammar.

Across simulated e-children, the null-subject parameter first rises and later falls back toward the obligatory-subject target. This matches O&H's hypothesis. The work offers a template for combining performance factors with competence learning in acquisition models.

Key points

  • The empirical basis is O&H's Experiment 2, a truth-value judgment task with 30 Los Angeles children in three age bands (2;6–2;11, 3;0–3;5, 3;6–3;11). Each child judged 24 items, 8 of them in the NS condition.
  • Children gave the adult imperative reading about 40% of the time in the youngest group and 64% in the middle group. The authors adopt roughly 90% for the oldest group, estimated from O&H's figure.
  • IARC grows over age under two alternative curves, linear and logistic. In the input, imperatives make up 16% of sentences before age 2;0 and 18% afterwards. With probability 1−IARC, an imperative is fed to the learner as a declarative.
  • There were 100 SSVL e-children per growth function. SSVL is built on a reward-only VL using the Bush–Mosteller linear update. It uses two learning rates, an aggressive 2×10⁻⁴ and a conservative 5×10⁻⁶, which the authors found by trial and error.
  • The NS weight starts at 0.5, climbs to about 0.8, and then declines, with a small bump near 2;0 caused by the increase in imperatives. A noiseless control converges well before 2;0, so it shows no NS stage at all.

Interesting ideas

  • I1. Model a processing limitation as its own time-varying coefficient that distorts what the learner receives, kept separate from the learner's grammar; this lets you test how input misreading shapes learned competence.
  • I2. Before rewarding a permissive hypothesis for explaining a sentence, check whether a stricter hypothesis also explains it, and prefer the stricter one; this guards against overgeneral conclusions that positive data alone can never refute.
  • I3. Run a noise-free control alongside the noisy simulation, so the effect being explained is shown to come from the injected factor rather than from the learner itself.
  • I4. Subjectless utterances are ambiguous in force (command vs. statement), and getting this wrong changes what is learned downstream; making the illocution choice an explicit, scored decision helps.
  • I5. Simulate a whole population of learners with different trajectories and report the fastest, slowest and median, instead of a single run.

For Mova

How these ideas could serve Mova, our small open language model — its files, stages and tests:

  • I1 (learn): Give each silver-data witness an explicit, possibly drifting error coefficient. Simulate how that noise propagates into trained parser weights before trusting the agreed labels.
  • I2 (learn): When self-inducing repair rules, test whether a narrower rule already covers the triggering cases. Adopt the broader rule only when the narrow one fails.
  • I3 (learn): Pair each evaluation with a clean-input or ablated-noise run as a negative control. Any gain or error pattern can then be attributed to the specific factor tested.
  • I4 (frames): In speech-act detection, mark subjectless clauses as imperative candidates with an implicit addressee as agent. Record the choice and its confidence so frames and later steps can explain it.
  • I5 (learn): Train several seeds or data orders and report spread with best, worst and median curves. This exposes how unstable a training recipe is before adopting it.

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, topics and concepts

📚 Shelves 4The field of science this work belongs to (OpenAlex fields).Arts 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.English · 2,379 works

Cite

Cite as Soumik Dey and William Sakas. 2025. Performance and competence intertwined: A computational model of the Null Subject stage in English-speaking children. Proceedings of the Second International Workshop on Construction Grammars and NLP.

The paper itself, not this page. BibTeX from the library's record of the work.

@inproceedings{dey2025performance,
  title = {{Performance and competence intertwined: A computational model of the Null Subject stage in English-speaking children}},
  author = {Soumik Dey and William Sakas},
  booktitle = {Proceedings of the Second International Workshop on Construction Grammars and NLP},
  year = {2025},
  url = {https://aclanthology.org/2025.cxgsnlp-1.10/}
}

References

42 references, 0 of them with a card of ours. ↗ leads to the work, «find» searches for one we have no id for yet.

  1. Katherine Finn Davis, Kathy P. Parker, and Gary L. Montgomery. 2004. Sleep in infants and young children: Part one: normal sleep. Journal of Pediatric Health Care, 18(2):65–71. find
  2. Robert C. Berwick. 1985. The Acquisition of Syntactic Knowledge. The MIT Press. find
  3. Paul Bloom. 1990. Subjectless sentences in child language. Linguistic Inquiry, 21(4):491–504. find
  4. Robert R. Bush and Frederick Mosteller. 1955. Stochastic Models for Learning. Wiley, New York, NY. find
  5. Noam Chomsky. 1955. The Logical Structure of Linguistic Theory. Plenum Press and University of Chicago Press., New York, NY and Chicago, IL. find
  6. Noam Chomsky. 1965. Aspects of the theory of syntax. MIT Press, Cambridge, MA. find
  7. Noam Chomsky. 1981. Lectures on government and binding. Foris, Dordrecht, Netherlands. find
  8. Noam Chomsky. 1986. Knowledge of Language: Its Nature, Origin, and Use. Praeger, New York, NY. find
  9. Robin Clark. 1992. The selection of syntactic knowledge. Language Acquisition, 2(2):83–149. find
  10. Robin Clark and Ian Roberts. 1993. A computational model of language learnability and language change. Linguistic Inquiry, 24(2):299–345. find
  11. Stephen Crain and Janet Dean Fodor. 1993. The acquisition of structural restrictions on anaphora. In Language and Cognition: A Developmental Perspective, volume 16, pages 141–171. Ablex. find
  12. Stephen Crain and Cecile McKee. 1985. Acquisition of structural restrictions on anaphora. In NELS, volume 16, pages 94–110. University of Massachusetts, Graduate Linguistic Student Association. find
  13. Viviane Déprez and Amy Pierce. 1993. Negation and functional projections in early grammar. Linguistic Inquiry, 24(1):25–67. find
  14. Janet Dean Fodor. 1998. Unambiguous triggers. Linguistic Inquiry, 29(1):1–36. Type: Journal Article. find
  15. Janet Dean Fodor and William G. Sakas. 2005. The subset principle in syntax: costs of compliance. Journal of Linguistics, 41(3):513–569. find
  16. Lila R Gleitman, Elissa L Newport, and Henry Gleitman. 1984. The current status of the motherese hypothesis. Journal of child language, 11(1):43–79. find
  17. Luigi Rizzi. 2005a. Grammatically-based target-inconsistencies in child language. In Proceedings of the Inaugural Conference of GALANA. UCON-N/MIT Working papers in Linguistics. find
  18. Luigi Rizzi. 2005b. On the grammatical basis of language development: A case study. In G. Cinque and R. Kayne, editors, The Oxford handbook of comparative syntax, pages 70–109. Oxford University Press. find
  19. E Mark Gold. 1967. Language identification in the limit. Information and Control, 10(5):447–474. find
  20. William Gregory Sakas. 2003. A word-order database for testing computational models of language acquisition. In Proceedings of the 41st Annual Meeting of the Association for Computational Linguistics, pages 415–422, Sapporo, Japan. Association for Computational Linguistics. find
  21. Betty Hart and Todd R Risley. 1995. Meaningful differences in the everyday experience of young American children. Paul H Brookes Publishing. find
  22. William Gregory Sakas and Janet Dean Fodor. 2011. Generating CoLAG languages using the “supergrammar.”. Technical report. find
  23. Betty Hart and Todd R Risley. 2003. The early catastrophe: The 30 million word gap by age 3. American educator, 27(1):4–9. find
  24. William Gregory Sakas and Janet Dean Fodor. 2012. Disambiguating syntactic triggers. Language Acquisition, 19(2):83–143. find
  25. Katherine Howitt, Soumik Dey, and William Gregory Sakas. 2021. Gradual syntactic triggering: The gradient parameter hypothesis. Language Acquisition, 28(1):65–96. find
  26. William Gregory Sakas, Charles Yang, and Robert Berwick. 2017. Parameter setting is feasible. Linguistic Analysis, 41:391–408. find
  27. Eric H Lenneberg. 1967. The biological foundations of language. Hospital Practice, 2(12):59–67. find
  28. Brian MacWhinney. 2000. The CHILDES Project: Tools for Analyzing Talk. Third Edition. Lawrence Erlbaum Associates, Mahwah, NJ. find
  29. M. Rita Manzini and Kenneth Wexler. 1987. Parameters, binding theory, and learnability. Linguistic Inquiry, 18(3):413–444. find
  30. Elissa L. Newport, Henry Gleitman, and Lila R. Gleitman. 1977. Mother, I’d rather do it myself: Some effects and non effects of maternal speech style. In Catherine Elizabeth Snow and Charles Albert Ferguson, editors, The acquisition of syntax, page 109–149. Cambridge University Press. find
  31. Robyn Orfitelli and Nina Hyams. 2008. An experimental study of children’s comprehension of null subjects: Implications for grammatical/performance accounts. In Proceedings of the 32nd Annual BU-CLD, volume 2, pages 335–346. find
  32. Robyn Orfitelli and Nina Hyams. 2012. Children’s grammar of null subjects: Evidence from comprehension. Linguistic Inquiry, 43(4):563–590. find
  33. Lisa Pearl and Jon Sprouse. 2021. The acquisition of linking theories: A Tolerance and Sufficiency Principle approach to deriving UTAH and rUTAH. Language Acquisition, 28(3):294–325. find
  34. Kenneth Jerold Straus. 2008. Validations of a Probabilistic Model of Language Learning. Ph.D. thesis, Department of Mathematics, Northeastern University. find
  35. Virginia Valian. 1990. Null subjects: A problem for parameter-setting models of language acquisition. Cognition, 35(2):105–122. find
  36. Virginia Valian. 1991. Syntactic subjects in the early speech of american and italian children. Cognition, 40(1):21–81. find
  37. Leslie Gabriel Valiant. 1984. A theory of the learnable. Communications of the ACM, 27(11):1134–1142. find
  38. Qi Wang, Diane Lillo-Martin, Catherine T Best, and Andrea Levitt. 1992. Null subject versus null object: Some evidence from the acquisition of chinese and english. Language acquisition, 2(3):221–254. find
  39. Charles Yang. 2002a. Knowledge and Learning in Natural Language. Oxford University Press, Oxford, UK. find
  40. Charles Yang. 2002b. Knowledge and Learning in Natural Language. Oxford University Press, Oxford, UK. find
  41. Charles Yang. 2012. Computational models of syntactic acquisition. WIREs Cognitive Science, 3(2):205– 213. find
  42. Wilder Penfield and Lamar Roberts. 1959. Speech and brain mechanisms. Princeton University Press. find

Updated

↑ back to top

This page in Markdown → https://md.research.elamur.ai/free/en/papers/dey-2025-performance-competence-intertwined-computational-model/