PAC: 1,000-10,000 Polish Abusive Contract Clauses for Classification

PAC (Polish Abusive Clauses) is a text classification dataset containing between 1,000 and 10,000 Polish contract clauses annotated for unfair or abusive statements. Created by researcher laugustyniak and updated in March 2026, the collection utilizes hired annotators to identify problematic legal language in consumer agreements.

Use Cases

• Multi-class classification of Polish legal text to identify unfair contract terms
• Training NLP models to detect abusive clauses in consumer Terms and Conditions
• Legal tech research into automated compliance checking for Polish language documents

Strengths

• 1,000 to 10,000 annotated records
• Expert-labeled by hired annotators rather than automated heuristics
• Supports multi-class classification for specific legal violations

Limitations

• Monolingual Polish coverage restricts use to a single language domain
• Size is limited to a maximum of 10,000 records
• Context is restricted to consumer law and may not generalize to commercial or international law

Provenance

Source

laugustyniak via Hugging Face

Collection Method

Annotated by hired annotators

Freshness

Last updated March 2026 according to platform metadata.

Geography

Poland