UD-CHILDES-BG: a dependency treebank of Bulgarian child and child-directed speech

2026-06-25 · Source: Paper Index on ACL Anthology · Field: Technology & Digital — Artificial Intelligence & Machine Learning · Depth: Expert, short

Summary

UD-CHILDES-BG is a newly introduced, manually corrected Universal Dependencies treebank for Bulgarian child and child-directed speech. This resource comprises 4,338 dependency parses, representing 10% of the original CHILDES-BG corpus, with 14% of these parses double-annotated for quality. Evaluation showed inter-annotator agreement (UAS/LAS) of 91.71/86.12 for child-directed speech (CDS) and 88.14/81.40 for child speech (CS). Parser performance on this corrected data reached 92.70/85.54 for CDS and 90.97/81.52 for CS. These scores are slightly lower than the 93.37/90.21 reported for adult written language, indicating that child and child-directed speech present unique challenges for dependency annotation and parsing, particularly due to discourse-related structures less prevalent in adult text.

Key takeaway

For NLP Engineers developing models for child or developmental language data, you should recognize that standard dependency parsers trained on adult written language will likely exhibit reduced performance. Your models must account for the distinct discourse-related structures prevalent in child and child-directed speech. Utilize specialized resources like UD-CHILDES-BG to fine-tune or train parsers, ensuring better accuracy and robustness for these challenging linguistic domains.

Key insights

UD-CHILDES-BG reveals unique challenges in parsing child and child-directed speech, especially discourse structures, compared to adult language.

Principles

• Child and child-directed speech parsing is more complex.
• Discourse structures pose unique annotation challenges.
• Developmental language data needs specialized treebanks.

Method

The method involves manually correcting 4,338 dependency parses, double-annotating 14% for agreement evaluation, and systematically analyzing parser errors on child and child-directed speech data.

In practice

• Utilize UD-CHILDES-BG for Bulgarian child NLP.
• Expect lower parser accuracy on child speech.
• Prioritize discourse structures in child language models.

Topics

• Universal Dependencies
• Bulgarian Language
• Child Language
• Dependency Parsing
• Linguistic Annotation
• Treebank

Best for: Research Scientist, AI Scientist, NLP Engineer

Related on AIssential

• A UD Parser to the Rescue: A Method for Bringing a Classical Annotated Corpus to Life Again — A knowledge-rich parsing method can effectively update classical linguistic corpora to modern annotation standards.
• AfriSUD: A Dependency Treebank Collection for Evaluating Models on African Languages — AfriSUD provides the first large-scale, native-speaker verified dependency treebanks for nine diverse African languages, revealing a significant syntax gap in current NLP models.
• Parsing English in 500 Lines of Python — Transition-based dependency parsing offers a breakthrough in NLU with a concise, dependency-free Python implementation.
• Parsing Nheengatu: Performance Gains for a Brazilian Indigenous Universal Dependencies Treebank — Corpus expansion and standardized annotation improve parsing accuracy and reduce variance for low-resource languages.
• A Scalable Tool for Measuring Manner and Result Verbs in Developmental Language Research — A RoBERTa classifier, trained on LLM-generated annotations, accurately distinguishes manner and result verbs at scale.
• Grammar as an Injectable: A Trojan Horse to NLP | Towards Data Science — CCG offers a non-statistical, algebraic framework for NLP, treating words as functions to ensure grammatical consistency and enhance interpretability.

Open in AIssential →

Editorial summary, takeaway, and curation by AIssential. Original article published by Paper Index on ACL Anthology.