legal_contract_named_entity_recognizer

legal_contract_named_entity_recognizer

Overview

This model is a BERT-based Token Classifier fine-tuned for the Legal domain. It automatically extracts key entities from commercial contracts, including the parties involved, effective dates, governing jurisdictions, and financial amounts.

Model Architecture

The model uses a BERT-Large backbone with a token-level classification head.

• Tagging Scheme: Follows the BIO (Beginning, Inside, Outside) format.
• Contextual Embeddings: Captures the dense semantic relationships between legal definitions (e.g., distinguishing between a "Notice Date" and an "Effective Date").
• Fine-tuning: Trained on the CUAD (Contract Understanding Atticus Dataset) and proprietary legal corpora.

Intended Use

• Contract Lifecycle Management (CLM): Automating the extraction of metadata for digital repositories.
• Due Diligence: Rapidly identifying governing laws and liability amounts across thousands of merger documents.
• Regulatory Compliance: Checking for the presence of specific mandatory parties or dates in financial agreements.

Limitations

• Legalese Variation: Older or highly non-standard contract formats may result in lower entity recall.
• Nested Entities: Does not support hierarchical or overlapping entities (e.g., an "Amount" inside a "Payment Clause").
• OCR Errors: Performance is highly dependent on the quality of the text; poorly scanned PDFs with OCR noise will degrade accuracy.