UD-English-CHILDES: A Collected Resource of Gold and Silver Universal Dependencies Trees for Child Language Interactions

What it is about

The authors collect several earlier dependency-annotated CHILDES corpora, harmonise them to one set of UD v2 principles and release the first official UD treebank of child-adult speech. The gold part has about 48k sentences (236k tokens) from 11 children and their caregivers. About 8,000 corrections were made, many because automatic UPOS tags did not match the dependency labels. A further 1M sentences are silver, parsed with Stanza. Stanza's LAS on the gold data is 83.3, which is the baseline the silver quality is judged by. The contribution is the resource and its validation, not a new parser.

Ideas

• I1. annot: check the silver quality of the teacher pipeline on the gold part of this treebank before any silver is used for training: LAS and UAS per child, with the gold trees as truth (test only).
• I2. en/src/convert.rs: a converter from the UD-English-CHILDES files into the mova dialect, checked by the UD validator and by the count of UPOS and deprel mismatches per 1k sentences.
• I3. en/src/ud.rs backoff tables: count UPOS-deprel mismatches in the silver output of en on CHILDES utterances; the paper's main correction class (en ud-eval on gold).

For Mova

This is the reference CHILDES treebank in UD, and Mova's CHILDES source is the one it lists for training. Before the silver part is used, the paper's quality check should be repeated with Mova's own parser. Relevant chapters are languages/en/10-word-classes (UPOS and the UPOS-deprel link) and 20-syntax-overview. Measure first the UPOS-deprel mismatch rate in en output on the gold part, with the split of this paper held out.

Topics: Universal Dependencies

Cite

The paper itself, not this card. BibTeX from the library's record of the work. arXiv 2504.20304

@inproceedings{yang2025ud,
  title = {{UD-English-CHILDES: A Collected Resource of Gold and Silver Universal Dependencies Trees for Child Language Interactions}},
  author = {Yang, Xiulin and Ju, Zhuoxuan and Bu, Lanni and Liu, Zoey and Schneider, Nathan},
  booktitle = {Proceedings of the Eighth Workshop on Universal Dependencies (UDW, SyntaxFest 2025)},
  year = {2025},
  eprint = {2504.20304},
  archivePrefix = {arXiv},
  url = {https://arxiv.org/abs/2504.20304}
}

References

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

1. Omri Abend, Tom Kwiatkowski, Nathaniel J. Smith, Sharon Goldwater, and Mark Steedman. 2017. Bootstrapping language acquisition. Cognition, 164:116– 143. find
2. Ruth A. Berman. 1990. On acquiring an (S)VO language: subjectless sentences in children’s Hebrew. Linguistics, 28(6):1135–1166. find

This page in Markdown → https://md.mova.elamur.ai/free/en/docs/papers/ud-2025-yang-ud-english-childes-collected-resource/