Sanskrit OCR Post-Correction Benchmark with Classical Texts

A benchmark and dataset for post-OCR text correction in Sanskrit, featuring manually post-edited OCR data for Devanagari script texts. It includes train/validation/test splits with OCR text and corrected ground truth, plus an out-of-domain test set of 500 sentences. Source texts are from classical Sanskrit works including Brahmasutra Bhashyam, Grahalaghava, and Goladhyaya.

Use Cases

• Train sequence-to-sequence models to correct OCR text using the provided ground truth for Sanskrit Devanagari script.
• Benchmark model generalization on the out-of-domain test set of 500 sentences.
• Analyze OCR error patterns across different classical Sanskrit source texts like Brahmasutra Bhashyam and Goladhyaya.
• Develop post-editing tools for Sanskrit digitization projects using the manually corrected ground truth data.

Strengths

• Includes a dedicated out-of-domain test set of 500 sentences for evaluating model generalization.
• Data is manually post-edited, providing high-quality ground truth for OCR correction tasks.
• Source texts are from specific, named classical Sanskrit works, providing clear provenance.

Limitations

• Dataset size, row count, and column structure are unspecified, limiting assessment of scale and feature richness.
• Focus is limited to Sanskrit in Devanagari script, which may not generalize to other languages or scripts.
• The source texts are from classical works, potentially introducing a domain bias for modern Sanskrit applications.

Provenance

Source

Classical Sanskrit works including Brahmasutra Bhashyam, Grahalaghava, and Goladhyaya.

Collection Method

Manually post-edited OCR data.

Freshness

Last updated on 2025-10-11.