avishadilhara/sinhala-ocr-lk-acts-1010

🇱🇰 Sinhala OCR - Sri Lankan Acts Dataset

Dataset Description

This dataset contains 1,010 scanned document images of Sri Lankan legal acts (1980s-2010s) in Sinhala language with ground truth text annotations for Optical Character Recognition (OCR) training and evaluation.

Key Features

• —✅ High-quality scanned document images
• —✅ Professionally corrected ground truth text
• —✅ Year-wise metadata for temporal analysis
• —✅ Pre-split into train/eval/test sets
• —✅ Suitable for document OCR and layout analysis

Dataset Summary

Dataset Structure

Data Splits

Data Fields

• —`image` (PIL.Image): Scanned page image in PNG format
• —`text` (string): Ground truth Sinhala text (OCR target)
• —`year` (string): Year of the document (1980-2019)
• —`doc_id` (string): Document identifier
• —`page_num` (string): Page number within document
• —`filename` (string): Original filename for reference

Year Distribution (Training Set)

• —1981: 76 samples
• —1982: 26 samples
• —1983: 10 samples
• —1984: 7 samples
• —1985: 7 samples
• —1986: 16 samples
• —1987: 43 samples
• —1988: 96 samples
• —1989: 8 samples
• —2000: 57 samples
• —2001: 22 samples
• —2002: 13 samples
• —2003: 14 samples
• —2004: 12 samples
• —2005: 16 samples
• —2006: 13 samples
• —2007: 23 samples
• —2008: 22 samples
• —2009: 23 samples
• —2010: 29 samples
• —2011: 70 samples
• —2012: 9 samples
• —2013: 10 samples
• —2014: 13 samples
• —2015: 9 samples
• —2016: 13 samples
• —2017: 20 samples
• —2018: 14 samples
• —2019: 16 samples

Applications This dataset is suitable for:

📄 Document OCR: Training models for Sinhala document recognition

🔍 Layout Analysis: Understanding document structure

📊 Temporal Analysis: Studying OCR performance across decades

🌐 Low-resource NLP: Expanding Sinhala language resources

🏛️ Digital Archiving: Digitizing historical legal documents

Citation If you use this dataset in your research, cite:

@misc{dilhara2026crosstemporalsinhalaocrpagelevel,
      title={Cross-Temporal Sinhala OCR: Page-Level Adaptation and Diachronic Analysis}, 
      author={Avisha Dilhara and Nevidu Jayatilleke},
      year={2026},
      eprint={2606.29378},
      archivePrefix={arXiv},
      primaryClass={cs.CL},
      url={[https://arxiv.org/abs/2606.29378](https://arxiv.org/abs/2606.29378)}
}

This dataset is released under CC-BY-4.0 license.

Version History v1.0 (2026-01): Initial release with 1,010 samples

Pre-split into train/eval/test sets

Year-wise metadata included

Keywords: OCR, Sinhala, Sri Lanka, NLP, Computer Vision, Historical Documents, Legal Documents