U.S. Court Decision Embeddings from the Caselaw Access Project

Over 40 million digitized U.S. court decisions, representing 6.7 million cases, are available through this project. Harvard Law Library, in collaboration with Ravel Law, created this collection spanning 360 years of legal history. The dataset was last updated on the platform in March 2024.

Use Cases

• Semantic search for legal precedents using precomputed text embeddings of case decisions.
• Training or fine-tuning language models on domain-specific legal text from the case corpus.
• Analyzing temporal shifts in legal concepts by comparing embedding vectors across the 360-year date range.
• Clustering similar cases based on embedding similarity for legal topic modeling.

Strengths

• Contains over 40 million court decision documents.
• Covers a 360-year temporal range of U.S. legal history.
• Represents 6.7 million distinct legal cases.

Limitations

• The specific embedding model, dimensionality, and column structure are not described.
• Sample data and file formats are unavailable for inspection.
• Potential geographic bias is present as the data is exclusively U.S. court decisions.

Provenance

Source

Harvard Law Library and Ravel Law via the Caselaw Access Project (CAPAPI).

Collection Method

Digitization and processing of historical and modern court records.

Time Range

Approximately 360 years of coverage.

Freshness

null

Geography

United States (federal and state courts).