shekar-ai/SynPerForm-synthetic-persian-formality-pairs

SynPerForm

SynPerForm is a paired Persian dataset for formality style transfer. Each informal text is paired with a freely written formal rewrite that preserves its meaning without requiring lexical or structural equivalence. The formal rewrites were generated with OpenAI GPT-5.6 Luna.

Columns

• —Informal: the original informal Persian text.
• —Formal: a free formal rewrite that preserves the original meaning.

Intended uses

• —Formality style transfer: train or evaluate systems that freely rewrite informal Persian in a formal style while preserving meaning.
• —Text classification: use texts from the Informal and Formal columns as examples of informal and formal language, respectively.
• —Representation learning: learn Persian representations from semantically aligned pairs that express similar meanings in different formality styles.

Splits

• —Train: 241,692 rows
• —Validation: 7,500 rows
• —Test: 7,500 rows