---
title: "Making Synthetic Questions More Child-Directed: Prompting and"
description: "Child-directed speech (CDS) can help small language models learn, but real CDS corpora are small, so researchers try to… 33 references."
authors: ["Whitney Poh", "Michael Tombolini", "Libby Barak"]
year: 2026
venue: "Proceedings of the 1st Workshop on Computational Developmental Linguistics (CDL)"
source: "https://aclanthology.org/2026.cdl-1.18/"
doi: "10.18653/v1/2026.cdl-1.18"
licence: "CC BY 4.0 (ACL Anthology)"
licence_url: "https://creativecommons.org/licenses/by/4.0/"
summary_licence: "Elamur Cosmic License"
canonical: "https://research.elamur.ai/free/en/papers/poh-2026-making-synthetic-questions-more-child/"
---
[Elamur Research](https://md.research.elamur.ai/free/en/) · [this page in HTML](https://research.elamur.ai/free/en/papers/poh-2026-making-synthetic-questions-more-child/)

[home](https://md.research.elamur.ai/free/en/) / [papers](https://md.research.elamur.ai/free/en/papers/) / [2026](https://md.research.elamur.ai/free/en/papers/2026/page/35/) / Making Synthetic Questions More Child-Directed: Prompting and Sampling Effects

# Making Synthetic Questions More Child-Directed: Prompting and Sampling Effects

**Authors:** [Whitney Poh](https://md.research.elamur.ai/free/en/authors/whitney-poh/), [Michael Tombolini](https://md.research.elamur.ai/free/en/authors/michael-tombolini/) and [Libby Barak](https://md.research.elamur.ai/free/en/authors/libby-barak/) · **Year:** 2026 · **Venue:** [Proceedings of the 1st Workshop on Computational Developmental Linguistics (CDL)](https://md.research.elamur.ai/free/en/venues/cdl/) ([CDL 2026](https://md.research.elamur.ai/free/en/venues/cdl/2026/)) · **Publisher:** [Association for Computational Linguistics](https://md.research.elamur.ai/free/en/publishers/association-for-computational-linguistics/) · **Licence:** [CC BY 4.0 (ACL Anthology)](https://creativecommons.org/licenses/by/4.0/)

**Links:** [Publisher's page](https://aclanthology.org/2026.cdl-1.18/) · [PDF](https://aclanthology.org/2026.cdl-1.18.pdf) · [DOI 10.18653/v1/2026.cdl-1.18](https://doi.org/10.18653/v1/2026.cdl-1.18) · [OpenAlex](https://openalex.org/W7166834311)

**Citations:** 0 citations (OpenAlex, 2026-10-14)

## Summary
Child-directed speech (CDS) can help small language models learn, but real CDS corpora are small, so researchers try to synthesize CDS-like input. This paper revisits an earlier attempt (Poh et al., 2025) to add caregiver-style questions to BabyLM training text, which had yielded questions that were too long and structurally unlike CDS. The authors used GPT-5-mini to insert questions into the non-CHILDES parts of the BabyLM 2025 10M corpus with two revised prompts. One asks for simple caregiver questions followed by answers. The other asks for brief pragmatic questions in any grammatical form. They down-sampled the data in several ways (random, MLU-based, question-type-based, and an MLU-balanced ablation), trained GPT-Wee models on about 14M words per condition, and evaluated them on BLiMP and three GLUE tasks. Describing communicative intent rather than linguistic specifics produced questions much closer to CDS in length and type distribution, and the choice of prompt mattered more than the sampling method. Downstream scores improved over the earlier synthetic data but did not consistently beat the unmodified baseline, so the value of questions as training data remains open.

## Key points
- The earlier synthetic questions had a mean MLU of 7.82 and 33.40% non-wh/non-yes-no questions. Real CDS has a mean MLU of 4.92 and 48.49% such questions. Prompt 2 lowered MLU most strongly, especially for Gutenberg, OpenSubtitles and Switchboard.
- The setup trained five GPT-Wee models per condition on the questioned 10M BabyLM data plus the full CHILDES set, about 14M words in total. Training used a learning rate of 5e-4, batch size 32 and 50,000 steps.
- On BLiMP, Prompt 2 Random was the best non-ablation model at 61.0% (std 0.005). Prompt 1 Balanced reached 61.7, nearly equal to the baseline. The balanced samples also had the lowest question MLU, 4.916.
- On GLUE, the baseline led on MRPC and STS-B, while Prompt 2 Random scored highest on RTE.
- Code and models are released at github.com/NLPlabMSU/BabyLM_Questions. Limitations include some topics that are not child-appropriate, a single tested architecture, and generated answers that were left out of the analysis.

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

## Institutions

- [Montclair State University](https://md.research.elamur.ai/free/en/universities/montclair-state-university/) — University, Montclair, United States · 5 works · 94 citations

📚 Shelves (6): [🖥️ Computer Science](https://md.research.elamur.ai/free/en/shelves/computer-science/) · 16,667 works · [🤖 Artificial Intelligence](https://md.research.elamur.ai/free/en/shelves/computer-science/artificial-intelligence/) · 16,194 works · [🎭 Arts and Humanities](https://md.research.elamur.ai/free/en/shelves/arts-and-humanities/) · 16,039 works · [🗣️ Language and Linguistics](https://md.research.elamur.ai/free/en/shelves/arts-and-humanities/language-and-linguistics/) · 15,891 works · [💭 Psychology](https://md.research.elamur.ai/free/en/shelves/psychology/) · 1,552 works · [💭 Developmental and Educational Psychology](https://md.research.elamur.ai/free/en/shelves/psychology/developmental-and-educational-psychology/) · 499 works

🧭 Topics (2): [🧠 Language models & pretraining](https://md.research.elamur.ai/free/en/topics/language-models/) · 3,320 works · [🧒 Psycholinguistics & language acquisition](https://md.research.elamur.ai/free/en/topics/psycholinguistics/) · 102 works

## Cite

Cite as: Whitney Poh, Michael Tombolini and Libby Barak. 2026. Making Synthetic Questions More Child-Directed: Prompting and Sampling Effects. Proceedings of the 1st Workshop on Computational Developmental Linguistics (CDL).

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

```bibtex
@inproceedings{poh2026making,
  title = {{Making Synthetic Questions More Child-Directed: Prompting and Sampling Effects}},
  author = {Whitney Poh, Michael Tombolini and Libby Barak},
  booktitle = {Proceedings of the 1st Workshop on Computational Developmental Linguistics (CDL)},
  year = {2026},
  doi = {10.18653/v1/2026.cdl-1.18},
  url = {https://doi.org/10.18653/v1/2026.cdl-1.18}
}
```

- [← previous Making Jobs Accessible through AI-supported Easy Language Translation](https://md.research.elamur.ai/free/en/papers/merkel-2026-making-jobs-accessible-through-ai/)
- [next → MameLoshnLM: Yiddish Language Model and Evaluation Benchmark](https://md.research.elamur.ai/free/en/papers/katz-2026-mameloshnlm-yiddish-language-model-evaluation/)
- [↑ back to papers of 2026](https://md.research.elamur.ai/free/en/papers/2026/page/35/)

## References

33 references, 15 of them with a card of ours. ↗ leads to the work, «find» searches for one we have no id for yet.

1. Yoshua Bengio, Jérôme Louradour, Ronan Collobert, and Jason Weston. 2009. [Curriculum learning](https://md.research.elamur.ai/free/en/papers/bengio-2009-curriculum-learning/). In Proceedings of the 26th annual international conference on machine learning, pages 41–48.
2. BNC Consortium. 2007. The british national corpus, xml edition. [find](https://scholar.google.com/scholar?q=BNC+Consortium%2E+2007%2E+The+british+national+corpus%2C+xml+edition%2E)
3. Peter Brodsky and Heidi Waterfall. 2007. Characterizing motherese: On the computational structure of child-directed language. In Proceedings of the annual meeting of the cognitive science society, volume 29. [find](https://scholar.google.com/scholar?q=Peter+Brodsky+and+Heidi+Waterfall%2E+2007%2E+Characterizing+motherese%3A+On+the+computational+structure+of+child%2Ddirected+language%2E+In+Proceedings+of+the+annual+meeting+of+the+cognitive+science+society%2C+vol)
4. Bastian Bunzeck and Sina Zarrieß. 2023. GPT-wee: How small can a small language model really get? In Proceedings of the BabyLM Challenge at the 27th Conference on Computational Natural Language Learning, pages 35–46, Singapore. Association for Computational Linguistics. [find](https://scholar.google.com/scholar?q=Bastian+Bunzeck+and+Sina+Zarrie%C3%9F%2E+2023%2E+GPT%2Dwee%3A+How+small+can+a+small+language+model+really+get%3F+In+Proceedings+of+the+BabyLM+Challenge+at+the+27th+Conference+on+Computational+Natural+Language+Learni)
5. William B. Dolan and Chris Brockett. 2005. [Automatically constructing a corpus of sentential paraphrases](https://md.research.elamur.ai/free/en/papers/dolan-2005-automatically-constructing-corpus-sentential-paraphrases/). In Proceedings of the Third International Workshop on Paraphrasing (IWP2005).
6. Ronen Eldan and Yuanzhi Li. 2023. [Tinystories: How small can language models be and still speak coherent english?](https://md.research.elamur.ai/free/en/papers/eldan-2023-tinystories-small-language-models-still/) Preprint, arXiv:2305.07759.
7. Steven Y. Feng, Noah D. Goodman, and Michael C. Frank. 2024. [Is child-directed speech effective training data for language models?](https://md.research.elamur.ai/free/en/papers/feng-2024-is-child-directed-speech-effective/) In Proceedings of the 2024 Conference on Empirical Methods in Natural Language Processing, pages 22055–22071, Miami, Florida, USA. Association for Computational Linguist…
8. Anita Gelboim and Elior Sulem. 2025. [TafBERTa: Learning grammatical rules from small-scale language acquisition data in Hebrew](https://md.research.elamur.ai/free/en/papers/gelboim-2025-tafberta-learning-grammatical-rules-small/). In Proceedings of the First BabyLM Workshop, pages 76–90, Suzhou, China. Association for Computational Linguistics.
9. Martin Gerlach and Francesc Font-Clos. 2020. A standardized project gutenberg corpus for statistical analysis of natural language and quantitative linguistics. Entropy, 22(1):126. [find](https://scholar.google.com/scholar?q=Martin+Gerlach+and+Francesc+Font%2DClos%2E+2020%2E+A+standardized+project+gutenberg+corpus+for+statistical+analysis+of+natural+language+and+quantitative+linguistics%2E+Entropy%2C+22%281%29%3A126%2E)
10. Daniel Cer, Mona Diab, Eneko Agirre, Iñigo Lopez-Gazpio, and Lucia Specia. 2017. [SemEval-2017 task 1: Semantic textual similarity multilingual and crosslingual focused evaluation](https://md.research.elamur.ai/free/en/papers/cer-2017-semeval-2017-task-1-semantic/). In Proceedings of the 11th International Workshop on Semantic Evaluation (SemEval-2017), pages 1–14, Vancouver, Canada.…
11. Lucas Charpentier, Leshem Choshen, Ryan Cotterell, Mustafa Omer Gul, Michael Hu, Jaap Jumelet, Tal Linzen, Jing Liu, Aaron Mueller, Candace Ross, Raj Sanjay Shah, Alex Warstadt, Ethan Wilcox, and Adina Williams. 2025a. Babylm turns 3: Call for papers for the 2025 babylm workshop. Preprint, arXiv:250… [find](https://scholar.google.com/scholar?q=Lucas+Charpentier%2C+Leshem+Choshen%2C+Ryan+Cotterell%2C+Mustafa+Omer+Gul%2C+Michael+Hu%2C+Jaap+Jumelet%2C+Tal+Linzen%2C+Jing+Liu%2C+Aaron+Mueller%2C+Candace+Ross%2C+Raj+Sanjay+Shah%2C+Alex+Warstadt%2C+Ethan+Wilcox%2C+and+Adin)
12. Lucas Charpentier, Leshem Choshen, Ryan Cotterell, Mustafa Omer Gul, Michael Y. Hu, Jing Liu, Jaap Jumelet, Tal Linzen, Aaron Mueller, Candace Ross, Raj Sanjay Shah, Alex Warstadt, Ethan Gotlieb Wilcox, and Adina Williams. 2025b. Findings of the third BabyLM challenge: Accelerating language modeling… [find](https://scholar.google.com/scholar?q=Lucas+Charpentier%2C+Leshem+Choshen%2C+Ryan+Cotterell%2C+Mustafa+Omer+Gul%2C+Michael+Y%2E+Hu%2C+Jing+Liu%2C+Jaap+Jumelet%2C+Tal+Linzen%2C+Aaron+Mueller%2C+Candace+Ross%2C+Raj+Sanjay+Shah%2C+Alex+Warstadt%2C+Ethan+Gotlieb+Wilco)
13. Richard Diehl Martinez, Zébulon Goriely, Hope Mc-Govern, Christopher Davis, Andrew Caines, Paula Akari Haga, Akiyo Fukatsu, Miyu Oba, Arianna Bisazza, and Yohei Oseki. 2024. BabyLM challenge: Exploring the effect of variation sets on language model training efficiency. In The 2nd BabyLM Challenge at… [find](https://scholar.google.com/scholar?q=Richard+Diehl+Martinez%2C+Z%C3%A9bulon+Goriely%2C+Hope+Mc%2DGovern%2C+Christopher+Davis%2C+Andrew+Caines%2C+Paula+Akari+Haga%2C+Akiyo+Fukatsu%2C+Miyu+Oba%2C+Arianna+Bisazza%2C+and+Yohei+Oseki%2E+2024%2E+BabyLM+challenge%3A+Explorin)
14. Michael Y. Hu, Aaron Mueller, Candace Ross, Adina Williams, Tal Linzen, Chengxu Zhuang, Ryan Cotterell, Leshem Choshen, Alex Warstadt, and Ethan Gotlieb Wilcox. 2024. Findings of the second BabyLM challenge: Sample-efficient pretraining on developmentally plausible corpora. In The 2nd BabyLM Challen… [find](https://scholar.google.com/scholar?q=Michael+Y%2E+Hu%2C+Aaron+Mueller%2C+Candace+Ross%2C+Adina+Williams%2C+Tal+Linzen%2C+Chengxu+Zhuang%2C+Ryan+Cotterell%2C+Leshem+Choshen%2C+Alex+Warstadt%2C+and+Ethan+Gotlieb+Wilcox%2E+2024%2E+Findings+of+the+second+BabyLM+cha)
15. Philip A. Huebner, Elior Sulem, Fisher Cynthia, and Dan Roth. 2021. [BabyBERTa: Learning more grammar with small-scale child-directed language](https://md.research.elamur.ai/free/en/papers/huebner-2021-babyberta-learning-more-grammar-small/). In Proceedings of the 25th Conference on Computational Natural Language Learning, pages 624–646, Online. Association for Computational Linguistics.
16. Jaap Jumelet, Lucas Charpentier, Michael Hu, and Jing Liu. 2025. Babylm_2025. OSF. [find](https://scholar.google.com/scholar?q=Jaap+Jumelet%2C+Lucas+Charpentier%2C+Michael+Hu%2C+and+Jing+Liu%2E+2025%2E+Babylm%5F2025%2E+OSF%2E)
17. Hector J Levesque, Ernest Davis, and Leora Morgenstern. 2011. The Winograd schema challenge. In AAAI Spring Symposium: Logical Formalizations of Commonsense Reasoning, volume 46, page 47. [find](https://scholar.google.com/scholar?q=Hector+J+Levesque%2C+Ernest+Davis%2C+and+Leora+Morgenstern%2E+2011%2E+The+Winograd+schema+challenge%2E+In+AAAI+Spring+Symposium%3A+Logical+Formalizations+of+Commonsense+Reasoning%2C+volume+46%2C+page+47%2E)
18. Pierre Lison and Jörg Tiedemann. 2016. [OpenSubtitles2016: Extracting large parallel corpora from movie and TV subtitles](https://md.research.elamur.ai/free/en/papers/lison-2016-opensubtitles2016-extracting-large-parallel-corpora/). In Proceedings of the Tenth International Conference on Language Resources and Evaluation (LREC’16), pages 923–929, Portorož, Slovenia. European Language Resources Association (EL…
19. Brian MacWhinney. 2000. The CHILDES project: The database, volume 2. Psychology Press. [find](https://scholar.google.com/scholar?q=Brian+MacWhinney%2E+2000%2E+The+CHILDES+project%3A+The+database%2C+volume+2%2E+Psychology+Press%2E)
20. Miyu Oba, Akari Haga, Akiyo Fukatsu, and Yohei Oseki. 2023. [BabyLM challenge: Curriculum learning based on sentence complexity approximating language acquisition](https://md.research.elamur.ai/free/en/papers/oba-2023-babylm-challenge-curriculum-learning-based/). In Proceedings of the BabyLM Challenge at the 27th Conference on Computational Natural Language Learning, pages 290–297, Singapore. Assoc…
21. Francesca Padovani, Bastian Bunzeck, Manar Ali, Omar Momen, Arianna Bisazza, Hendrik Buschmeier, and Sina Zarrieß. 2025a. Dialogue is not enough to make a communicative BabyLM (but neither is developmentally inspired reinforcement learning). In Proceedings of the First BabyLM Workshop, pages 421– 43… [find](https://scholar.google.com/scholar?q=Francesca+Padovani%2C+Bastian+Bunzeck%2C+Manar+Ali%2C+Omar+Momen%2C+Arianna+Bisazza%2C+Hendrik+Buschmeier%2C+and+Sina+Zarrie%C3%9F%2E+2025a%2E+Dialogue+is+not+enough+to+make+a+communicative+BabyLM+%28but+neither+is+developm)
22. Francesca Padovani, Jaap Jumelet, Yevgen Matusevych, and Arianna Bisazza. 2025b. [Child-directed language does not consistently boost syntax learning in language models](https://md.research.elamur.ai/free/en/papers/padovani-2025-child-directed-language-does-not/). In Proceedings of the 2025 Conference on Empirical Methods in Natural Language Processing, pages 19735–19756, Suzhou, China. Associ…
23. Whitney Poh, Michael Tombolini, and Libby Barak. 2025. [What did you say? generating child-directed speech questions to train LLMs](https://md.research.elamur.ai/free/en/papers/poh-2025-what-did-you-say-generating/). In Proceedings of the First BabyLM Workshop, pages 237–245, Suzhou, China. Association for Computational Linguistics.
24. Jessica F Schwab and Casey Lew-Williams. 2016. Repetition across successive sentences facilitates young children’s word learning. Developmental psychology, 52(6):879. [find](https://scholar.google.com/scholar?q=Jessica+F+Schwab+and+Casey+Lew%2DWilliams%2E+2016%2E+Repetition+across+successive+sentences+facilitates+young+children%E2%80%99s+word+learning%2E+Developmental+psychology%2C+52%286%29%3A879%2E)
25. Nikitas Theodoropoulos, Giorgos Filandrianos, Vassilis Lyberatos, Maria Lymperaiou, and Giorgos Stamou. 2024. BERTtime stories: Investigating the role of synthetic story data in language pre-training. In The 2nd BabyLM Challenge at the 28th Conference on Computational Natural Language Learning, page… [find](https://scholar.google.com/scholar?q=Nikitas+Theodoropoulos%2C+Giorgos+Filandrianos%2C+Vassilis+Lyberatos%2C+Maria+Lymperaiou%2C+and+Giorgos+Stamou%2E+2024%2E+BERTtime+stories%3A+Investigating+the+role+of+synthetic+story+data+in+language+pre%2Dtraining%2E)
26. Yulia Tsvetkov, Manaal Faruqui, Wang Ling, Brian MacWhinney, and Chris Dyer. 2016. Learning the curriculum with Bayesian optimization for task-specific word representation learning. In Proceedings of the 54th Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers), pa… [find](https://scholar.google.com/scholar?q=Yulia+Tsvetkov%2C+Manaal+Faruqui%2C+Wang+Ling%2C+Brian+MacWhinney%2C+and+Chris+Dyer%2E+2016%2E+Learning+the+curriculum+with+Bayesian+optimization+for+task%2Dspecific+word+representation+learning%2E+In+Proceedings+of+)
27. Alex Wang, Amanpreet Singh, Julian Michael, Felix Hill, Omer Levy, and Samuel R. Bowman. 2019. [Glue: A multi-task benchmark and analysis platform for natural language understanding](https://md.research.elamur.ai/free/en/papers/wang-2018-glue-multi-task-benchmark-analysis/). Preprint, arXiv:1804.07461.
28. Alex Warstadt, Aaron Mueller, Leshem Choshen, Ethan Wilcox, Chengxu Zhuang, Juan Ciro, Rafael Mosquera, 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 the… [find](https://scholar.google.com/scholar?q=Alex+Warstadt%2C+Aaron+Mueller%2C+Leshem+Choshen%2C+Ethan+Wilcox%2C+Chengxu+Zhuang%2C+Juan+Ciro%2C+Rafael+Mosquera%2C+Bhargavi+Paranjabe%2C+Adina+Williams%2C+Tal+Linzen%2C+and+Ryan+Cotterell%2E+2023%2E+Findings+of+the+BabyLM)
29. 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](https://md.research.elamur.ai/free/en/papers/warstadt-2020-blimp-benchmark-linguistic-minimal-pairs/). Transactions of the Association for Computational Linguistics, 8:377– 392.
30. Wikimedia. 2023. Simple english wikipedia dump. [https://dumps.wikimedia.org/simplewiki/](https://dumps.wikimedia.org/simplewiki/) 20230301/. Accessed: 2023-07-31. [↗](https://dumps.wikimedia.org/simplewiki/)
31. Yue Yu, Elizabeth Bonawitz, and Patrick Shafto. 2019. Pedagogical questions in parent–child conversations. Child development, 90(1):147–161. [find](https://scholar.google.com/scholar?q=Yue+Yu%2C+Elizabeth+Bonawitz%2C+and+Patrick+Shafto%2E+2019%2E+Pedagogical+questions+in+parent%E2%80%93child+conversations%2E+Child+development%2C+90%281%29%3A147%E2%80%93161%2E)
32. Aaditya Singh, Adam Fry, Adam Perelman, Adam Tart, Adi Ganesh, Ahmed El-Kishky, Aidan McLaughlin, Aiden Low, AJ Ostrow, Akhila Ananthram, Akshay Nathan, Alan Luo, Alec Helyar, Aleksander Madry, Aleksandr Efremov, Aleksandra Spyra, Alex Baker-Whitcomb, Alex Beutel, Alex Karpenko, and 465 others. 2025… — [OpenAI GPT-5 System Card](https://md.research.elamur.ai/free/en/papers/singh-2025-openai-gpt-5-system-card/)
33. Andreas Stolcke, Klaus Ries, Noah Coccaro, Elizabeth Shriberg, Rebecca Bates, Daniel Jurafsky, Paul Taylor, Rachel Martin, Carol Van Ess-Dykema, and Marie Meteer. 2000. [Dialogue act modeling for automatic tagging and recognition of conversational speech](https://md.research.elamur.ai/free/en/papers/stolcke-2000-dialogue-act-modeling-automatic-tagging/). Computational linguistics, 26(3):339–373.

*Updated 1 November 2026*