ParsProof: A Fine-Grained Benchmark for Persian Proofreading Tools

50 annotated sentences form a public subset of the ParsProof benchmark for evaluating Persian proofreading systems. The dataset features fine-grained annotations across 52 error types tailored to the linguistic properties of Persian. It was created by sbunlp and last updated on Hugging Face in February 2026.

Use Cases

• Benchmarking Persian proofreading systems based on the 52 annotated error types.
• Conducting detailed error-level analysis of Persian text, as mentioned in the description.
• Training or fine-tuning models for Persian grammatical error correction using the annotated sentences.

Strengths

• Features fine-grained annotations across 52 distinct error types.
• The benchmark is specifically tailored to the linguistic properties of the Persian language.
• Enables detailed error-level analysis, unlike many existing datasets.

Limitations

• The public subset comprises only 50 sentences; the full dataset size is unknown.
• Column-level documentation is absent; field semantics must be inferred after download.
• Row count is unknown, which may limit suitability assessment.

Provenance

Source

sbunlp on Hugging Face

Collection Method

Likely manually or semi-automatically annotated for proofreading errors.

Freshness

Last updated 2026-02-18 20:09:32; freshness should be verified.