Description
This model aims to detect License grants / permissions in agreements, provided by a Subject (PERMISSION_SUBJECT) to a Recipient (PERMISSION_INDIRECT_OBJECT). THe permission itself is in PERMISSION tag.
There is a lighter (non-transformer based) version of this model available as legner_grants_md.
Predicted Entities
PERMISSION, PERMISSION_SUBJECT, PERMISSION_OBJECT, PERMISSION_INDIRECT_OBJECT
How to use
documentAssembler = nlp.DocumentAssembler()\
.setInputCol("text")\
.setOutputCol("document")
tokenizer = nlp.Tokenizer()\
.setInputCols("document")\
.setOutputCol("token")
tokenClassifier = legal.BertForTokenClassification.pretrained("legner_bert_grants", "en", "legal/models")\
.setInputCols("token", "document")\
.setOutputCol("label")\
.setCaseSensitive(True)
pipeline = nlp.Pipeline(stages=[
documentAssembler,
tokenizer,
tokenClassifier
]
)
import pandas as pd
p_model = pipeline.fit(spark.createDataFrame(pd.DataFrame({'text': ['']})))
text = """Fox grants to Licensee a limited, exclusive (except as otherwise may be provided in this Agreement),
non-transferable (except as permitted in Paragraph 17(d)) right and license"""
res = p_model.transform(spark.createDataFrame([[text]]).toDF("text"))
from pyspark.sql import functions as F
res.select(F.explode(F.arrays_zip('token.result', 'label.result')).alias("cols")) \
.select(F.expr("cols['0']").alias("token"),
F.expr("cols['1']").alias("ner_label"))\
.show(20, truncate=100)
Results
+----------------+----------------------------+
| token| ner_label|
+----------------+----------------------------+
| Fox| B-PERMISSION_SUBJECT|
| grants| O|
| to| O|
| Licensee|B-PERMISSION_INDIRECT_OBJECT|
| a| O|
| limited| B-PERMISSION|
| ,| I-PERMISSION|
| exclusive| I-PERMISSION|
| (| I-PERMISSION|
| except| I-PERMISSION|
| as| I-PERMISSION|
| otherwise| I-PERMISSION|
| may| I-PERMISSION|
| be| I-PERMISSION|
| provided| I-PERMISSION|
| in| I-PERMISSION|
| this| I-PERMISSION|
| Agreement| I-PERMISSION|
| ),| I-PERMISSION|
|non-transferable| I-PERMISSION|
+----------------+----------------------------+
Model Information
| Model Name: | legner_bert_grants |
| Type: | legal |
| Compatibility: | Legal NLP 1.0.0+ |
| License: | Licensed |
| Edition: | Official |
| Input Labels: | [sentence, token] |
| Output Labels: | [ner] |
| Language: | en |
| Size: | 412.2 MB |
| Case sensitive: | true |
| Max sentence length: | 128 |
References
Manual annotations on CUAD dataset
Benchmarking
label precision recall f1-score support
B-PERMISSION 0.88 0.79 0.83 38
B-PERMISSION_INDIRECT_OBJECT 0.85 0.94 0.89 36
B-PERMISSION_SUBJECT 0.89 0.85 0.87 40
I-PERMISSION 0.80 0.69 0.74 342
O 0.94 0.97 0.95 1827
accuracy - - 0.92 2292
macro-avg 0.85 0.81 0.86 2292
weighted-avg 0.91 0.92 0.91 2292