BabyLM-community/babylm-zho

BabyLM Dataset

Dataset Description

This dataset is part of the BabyLM multilingual collection.
More information at: babylm.github.io/babybabellm

Dataset Summary

• Language: zho
• Script: Hani, Hans, Latn
• Tier: > 100M
• Byte Premium Factor: 0.935966
• Size (MB): 518.85
• Expected Size (MB): 508.23
• Number of Documents: 203,891
• Total Tokens: 137,835,046
• Tokenizer: Qwen/Qwen3-0.6B

Tokens Per Category

• child-available-speech: 98,731,442 tokens
• child-books: 15,976,688 tokens
• child-directed-speech: 9,636,590 tokens
• child-wiki: 453,092 tokens
• educational: 13,037,234 tokens

Tokens Per Group

• Transcription: 108,368,032 tokens
• Education: 13,037,234 tokens
• Books, Wiki, News: 16,429,780 tokens
• Subtitles: 0 tokens
• Padding: 0 tokens

Data Fields

• text: The document text
• doc-id: Unique identifier for the document
• category: Type of content (e.g., child-directed-speech, educational, etc.)
• data-source: Original source of the data
• script: Writing system used (ISO 15924)
• age-estimate: Target age or age range
• license: Data license
• misc: Additional metadata (JSON string)
• num-tokens: Number of tokens per item (based on white-space split)
• language: Language code (ISO 639-3)

Licensing Information

Please see license in individual documents

Data Sources & Attribution

Books

• Quangushi (full stories)
  • source: http://quangushi.com/
  • description: Collection of children’s stories
• GlotStoryBooks Collection
  • source: https://arxiv.org/abs/2310.16248
  • description: Collection of children storybooks

Educational material (exam style)

• GAOKAO
  • source: https://arxiv.org/abs/2305.12474
  • description: Dataset including subjective and objective questions from exams from 2010 to 2024
• CK-12
  • source: https://dl.acm.org/doi/10.1609/aaai.v38i17.29914
  • description: Dataset covering most comprehensive knowledge points in Chinese K12 field
• CSQ
  • source: https://figshare.com/articles/dataset/28667489
  • description: Chinese Science Question dataset
• FCGEC
  • source: https://aclanthology.org/2022.findings-emnlp.137/
  • description: A human-annotated corpus based on multi-choice grammatical error problems
• Primary school students’ compositions
  • source: https://journals.plos.org/plosone/article?id=10.1371/journal.pone.0309717
  • description: A hierarchical corpus of primary school students’ compositions

Reading comprehension

• CFT reading comprehension dataset
  • source: https://aclanthology.org/C16-1167/
  • description: Collection of children’s stories from a reading comprehension dataset
• CMRC-2019 reading comprehension dataset
  • source: https://aclanthology.org/2020.coling-main.589/
  • description: Collection of children’s stories from a reading comprehension dataset

Transcription

• ChildMandarin: A Comprehensive Mandarin Speech Dataset for Young Children Aged 3-5
  • source: https://aclanthology.org/2025.acl-long.614/
  • description: Transcriptions of children's conversations
• NaturalConv
  • source: http://ojs.aaai.org/index.php/AAAI/article/view/17649
  • description: A multi-turn topic-driven adult conversation dataset
• CHILDES
  • source: https://talkbank.org/childes/
  • description: Transcription of child directed language conversations

Wikis

• WikiJunior
  • source: https://en.wikibooks.org/wiki/Wikijunior:Languages/Mandarin_Chinese
  • description: Age-appropriate non-fiction books for children
• Wikibooks
  • source: https://en.wikibooks.org/wiki/Chinese_(Mandarin)
  • description: Age-appropriate non-fiction books for children

Data Curators

• Hai Hu
• Linyang He
• Siyuan Spng
• Ziyin Zhang

Comments

None