Modelpublic

thechancerylaneproject/legal-sustainability-ner

sourceHugging Facemitupdated 5mo agoView on Hugging Face

0likes

Model Card

Legal Sustainability NER

A bespoke spaCy named entity recognition model fine-tuned on legal text from The Chancery Lane Project (TCLP) — a collection of climate-focused contract clauses and glossary terms.

Fine-tuned on top of nlpaueb/legal-bert-base-uncased.

Used as part of the TCLP Knowledge Graph pipeline.

Entity labels

Label	Description
`Clause Name (TCLP-specific)`	A named TCLP clause (e.g. "Marni's Clause")
`Glossary Term (TCLP-specific)`	A named TCLP glossary term (e.g. "Net Zero Target")
`Concept`	A sustainability or legal concept (e.g. "carbon footprint", "force majeure")
`Defined Term`	A term defined within a contract (e.g. "Supplier", "Business Days")
`Party to Contract`	A contracting party (e.g. "Landlord", "Contractor")
`Org`	An organisation (e.g. "Science Based Targets Initiative")
`Standard`	A standard or framework (e.g. "ISO 27001", "GHG Protocol")
`Law`	A piece of legislation (e.g. "Data Protection Act 2018")
`Person`	A named individual
`Object`	A physical or tangible object
`Sector`	An industry or sector
`Other`	Entities that don't fit another category
`UNKNOWN`	Unclassifiable entities

Usage

python

import spacy
from huggingface_hub import snapshot_download

# Download model
model_path = snapshot_download("georgiaray/legal-sustainability-ner")

# Load
nlp = spacy.load(model_path + "/model-last")

# Run
doc = nlp("The Supplier must comply with the GHG Protocol and report Scope 1, 2 and 3 emissions.")
for ent in doc.ents:
    print(ent.text, "->", ent.label_)

Files

File	Description
`model-last/`	Latest model checkpoint
`entities.db`	SQLite database of ~2,200 hand-curated TCLP entity keywords, used for keyword matching alongside the NER model

Performance

~73% accuracy on entity type classification, evaluated against a hand-labelled dataset of TCLP entities. This is a subjective multi-label task where reasonable disagreement exists between labels (e.g. a term may be both a Defined Term and a Party to Contract).