Dataset Card for "Jawi-OCR-v1"

Jawi-OCR-v1

Dataset Summary

Jawi-OCR-v1 is a specialized dataset created for OCR (Optical Character Recognition) of historical Malay texts written in Jawi script (Arabic script adapted for Malay language). The dataset comprises sentences extracted from Warta Malaya, one of the pioneering independent Malay daily newspapers from the 1930s in Singapore. It includes both Jawi script and English text, reflecting the bilingual nature of historical Singaporean newspapers.

Supported Tasks and Leaderboards

• Tasks: OCR (Optical Character Recognition)
• Languages: Malay (Jawi script), English

Associated Models

• Model Name: mevsg/qwen-for-jawi-v1
• Model Type: OCR
• Performance Metrics:
  • Character Error Rate (CER): 8.66%
  • Word Error Rate (WER): 25.50%

Dataset Structure

• Data Instances: Each instance consists of an image and its corresponding UTF-8 text transcription
• Total Size: 1,129 examples
• Splits:
  • Training: 911 examples
  • Validation: 218 examples

Dataset Creation

Source Data

Text Source:

• Warta Malaya newspaper (1930s)
• More information about the source: Warta Malaya - National Library Board Singapore

Annotations

Annotation Process:

• Manual transcription of Jawi script into UTF-8 representation
• Annotated by research assistants at the National University of Singapore in August-September 2024
• Gold standard annotations with high quality control

Considerations for Using the Data

Social Impact of Dataset

This dataset contributes to the digital preservation and accessibility of historical Malay texts, supporting cultural heritage preservation and historical research in Southeast Asian studies.

Discussion of Biases

The dataset is sourced from a single newspaper from the 1930s, which may reflect the linguistic patterns and content typical of that specific period and publication.

Other Known Limitations

• Small dataset size (1,129 examples total)
• Limited to one historical source (Warta Malaya)
• Focuses specifically on newspaper text, which may not represent other types of Jawi writing

Additional Information

Dataset Curators

Research team at the National University of Singapore, lead by Miguel Escobar Varela.

Related Projects

• Jawi Transcription Project (https://fromthepage.com/mulaika/jawi-transcription-project/)
• Malay Concordance Project (https://mcp.anu.edu.au/Q/texts.html)

Note: While these related projects work with Jawi-to-Rumi (Romanized Malay) transliteration, this dataset uniquely provides UTF-8 representations of Jawi script.

Citation Information

If you use this dataset in your research, please cite it as:

@dataset{jawi_ocr_2024,
  title     = {Jawi-OCR-v1: A Gold Standard Dataset for Historical Malay OCR},
  author    = {[Miguel Escobar Varela]}, 
  year      = {2024},
  publisher = {HuggingFace},
  url       = {[https://huggingface.co/datasets/mevsg/Jawi-OCR-v1]},
  note      = {Dataset created at National University of Singapore for OCR of historical Malay newspapers}
}

Or in text format:

[Miguel Escobar Varela]. (2024). Jawi-OCR-v1: A Gold Standard Dataset for Historical Malay OCR [Dataset]. HuggingFace. [https://huggingface.co/datasets/mevsg/Jawi-OCR-v1]

Licensing Information

This dataset is released under a Creative Commons Attribution-NonCommercial-ShareAlike 4.0 International License.

This license:

• Allows others to share and adapt the dataset for non-commercial purposes
• Requires appropriate credit to the original creators
• Requires any derivative works to be shared under the same license terms
• Prohibits commercial use without explicit permission

For commercial use, please contact [contact information].

Terms of Use

1. Attribution — You must give appropriate credit to the National University of Singapore research team, provide a link to the license, and indicate if changes were made.
2. NonCommercial — You may not use the material for commercial purposes without explicit permission.
3. ShareAlike — If you remix, transform, or build upon the material, you must distribute your contributions under the same license as the original.