BabyLM-community/babylm-xho

BabyLM Dataset

Dataset Description

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

Dataset Summary

• Language: xho
• Script: Latn
• Tier: 1M
• Byte Premium Factor: 1.198782
• Size (MB): 6.52
• Expected Size (MB): 6.51
• Number of Documents: 2,548
• Total Tokens: 679,405
• Tokenizer: separate by whitespace

Tokens Per Category

• child-books: 98,144 tokens
• child-directed-speech: 74,950 tokens
• educational: 65,208 tokens
• padding-wikipedia: 387,662 tokens
• qed: 29,099 tokens
• simplified-text: 24,342 tokens

Tokens Per Group

• Transcription: 74,950 tokens
• Education: 65,208 tokens
• Books, Wiki, News: 98,144 tokens
• Subtitles: 29,099 tokens
• Padding: 412,004 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

• isiXhosa T2X
  • source: https://aclanthology.org/2024.lrec-main.1464/
  • description: Dataset curated in the study by Meyer and Buys, 2024

Educational

• QED (Afrikaans, isiXhosa, isiZulu)
  • source: https://aclanthology.org/L14-1675/
  • description: Dataset curated in the study by Abdelali et al., 2014
• High School Exam (all 5 South African languages)
  • source: https://openhumanitiesdata.metajnl.com/articles/10.5334/johd.108
  • description: Collection of high school exams curated by Sibeko and Zaanen, 2023

Wikis

• Wikipedia
  • source: n/a
  • description: Online encyclopedia

Data Curators

• Francois Meyer

Comments

None