Legal And Administrative Corpus For Language Modeling

The Pile Of Law is a curated corpus of legal and administrative data intended for analyzing data filtering norms and pretraining legal-domain language models. The dataset is monolingual (English) and originates from US sources. Its size is categorized as between 10 million and 100 million units.

Use Cases

• Pretrain a masked language model on US legal text using the 'masked language modeling' task tag
• Analyze data filtering norms across aggregated legal and administrative sources
• Fine-tune a language model for access-to-justice initiatives using legal-domain text

Strengths

• Corpus size is categorized as between 10 million and 100 million units
• Data is specifically curated for legal-domain language model pretraining
• Dataset is tagged for the 'fill-mask' task, indicating suitability for masked language modeling

Limitations

• Specific row count, column structure, and file formats are unknown
• Geographic coverage is limited to US sources, as indicated by the 'us' region tag
• The dataset's last update was in January 2023, which may limit temporal relevance

Provenance

Source

pile-of-law

Collection Method

Curated aggregation of legal and administrative data sources.

Freshness

Last updated on 2023-01-08.

Geography

United States (based on the 'us' region tag)