NUPunkt

NUPunkt is a specialized sentence boundary detection library optimized for legal documents. It achieves 91.1% precision while processing 10 million characters per second, providing a 29-32% improvement over general-purpose tools.

Key Innovations

• Legal-specific knowledge base with over 4,000 domain abbreviations
• Zero dependencies - pure Python implementation
• Exceptional performance - processes multi-million document collections in minutes
• High precision - critical for legal retrieval and analysis pipelines

Technical Details

NUPunkt handles the unique challenges of legal text:

• Complex citations (e.g., “See 15 U.S.C. § 78j(b).”)
• Hierarchical enumerations
• Multi-sentence quotations
• Latin phrases and specialized abbreviations

Applications

• Legal document retrieval systems
• E-discovery platforms
• Contract analysis pipelines
• Regulatory compliance tools

Related Work

• Precise Legal Sentence Boundary Detection for Retrieval at Scale: NUPunkt and CharBoundary
• Published in 2025 with Daniel Martin Katz and Jillian Bommarito
• Interactive demo at sentences.aleainstitute.ai