Persian Punctuation Restoration Dataset with 17 Million Samples

PersianPunc is a large-scale dataset for Persian punctuation restoration, containing 17 million token-level sequence labeling samples aggregated from 6 source corpora. It was created by MohammadJRanjbar and accepted at the EACL 2026 SilkRoad NLP Workshop.

Use Cases

• Train a sequence labeling model on 17 million samples for Persian punctuation restoration.
• Fine-tune a ParsBERT model for token-level punctuation prediction using the large-scale sample data.
• Benchmark punctuation restoration models for Persian text using the token-level sequence labeling task.
• Analyze punctuation patterns across the 6 aggregated source corpora included in the dataset.

Strengths

• 17 million samples provide a substantial foundation for model training.
• Data is aggregated from 6 distinct source corpora, offering potential diversity.
• Dataset is designed for token-level sequence labeling, a standard NLP task format.
• Peer-reviewed work accepted at the EACL 2026 SilkRoad NLP Workshop.

Limitations

• Specific column structure and features are not described, limiting understanding of data granularity.
• The raw description is brief, lacking details on label distribution, potential class imbalance, or data quality metrics.
• Geographic and temporal coverage of the source texts is unspecified.

Provenance

Source

MohammadJRanjbar via Hugging Face.

Collection Method

Aggregated from 6 source corpora; specific gathering method not detailed.

Freshness

Last updated on 2026-03-20.

Geography

Persian (Farsi) language data; specific geographic coverage unknown.