---
language: en
license: mit
tags:
- ner
- legal-nlp
- token-classification
- bert
---

# 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.