elenanereiss/bert-german-ler
Token Classification
•
Updated
•
103
•
3
Datasets:
Tasks:
Token Classification
Sub-tasks:
named-entity-recognition
Languages:
German
Size:
1M<n<10M
ArXiv:
DOI:
License:
Dataset Viewer
Full Screen Viewer
Full Screen
The viewer is disabled because this dataset repo requires arbitrary Python code execution. Please consider
removing the
loading script
and relying on
automated data support
(you can use
convert_to_parquet
from the datasets library). If this is not possible, please
open a discussion
for direct help.
Dataset Card for "German LER"
Dataset Summary
A dataset of Legal Documents from German federal court decisions for Named Entity Recognition. The dataset is human-annotated with 19 fine-grained entity classes. The dataset consists of approx. 67,000 sentences and contains 54,000 annotated entities. NER tags use the BIO tagging scheme.
The dataset includes two different versions of annotations, one with a set of 19 fine-grained semantic classes (ner_tags) and another one with a set of 7 coarse-grained classes (ner_coarse_tags). There are 53,632 annotated entities in total, the majority of which (74.34 %) are legal entities, the others are person, location and organization (25.66 %).
For more details see https://arxiv.org/pdf/2003.13016v1.pdf.
Supported Tasks and Leaderboards
- Tasks: Named Entity Recognition
- Leaderboards:
Languages
German
Dataset Structure
Data Instances
{
'id': '1',
'tokens': ['Eine', 'solchermaßen', 'verzögerte', 'oder', 'bewusst', 'eingesetzte', 'Verkettung', 'sachgrundloser', 'Befristungen', 'schließt', '§', '14', 'Abs.', '2', 'Satz', '2', 'TzBfG', 'aus', '.'],
'ner_tags': [38, 38, 38, 38, 38, 38, 38, 38, 38, 38, 3, 22, 22, 22, 22, 22, 22, 38, 38],
'ner_coarse_tags': [14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 2, 9, 9, 9, 9, 9, 9, 14, 14]
}
Data Fields
{
'id': Value(dtype='string', id=None),
'tokens': Sequence(feature=Value(dtype='string', id=None),
length=-1, id=None),
'ner_tags': Sequence(feature=ClassLabel(num_classes=39,
names=['B-AN',
'B-EUN',
'B-GRT',
'B-GS',
'B-INN',
'B-LD',
'B-LDS',
'B-LIT',
'B-MRK',
'B-ORG',
'B-PER',
'B-RR',
'B-RS',
'B-ST',
'B-STR',
'B-UN',
'B-VO',
'B-VS',
'B-VT',
'I-AN',
'I-EUN',
'I-GRT',
'I-GS',
'I-INN',
'I-LD',
'I-LDS',
'I-LIT',
'I-MRK',
'I-ORG',
'I-PER',
'I-RR',
'I-RS',
'I-ST',
'I-STR',
'I-UN',
'I-VO',
'I-VS',
'I-VT',
'O'],
id=None),
length=-1,
id=None),
'ner_coarse_tags': Sequence(feature=ClassLabel(num_classes=15,
names=['B-LIT',
'B-LOC',
'B-NRM',
'B-ORG',
'B-PER',
'B-REG',
'B-RS',
'I-LIT',
'I-LOC',
'I-NRM',
'I-ORG',
'I-PER',
'I-REG',
'I-RS',
'O'],
id=None),
length=-1,
id=None)
}
Data Splits
| train | validation | test | |
|---|---|---|---|
| Input Sentences | 53384 | 6666 | 6673 |
Dataset Creation
Curation Rationale
Documents in the legal domain contain multiple references to named entities, especially domain-specific named entities, i. e., jurisdictions, legal institutions, etc. Legal documents are unique and differ greatly from newspaper texts. On the one hand, the occurrence of general-domain named entities is relatively rare. On the other hand, in concrete applications, crucial domain-specific entities need to be identified in a reliable way, such as designations of legal norms and references to other legal documents (laws, ordinances, regulations, decisions, etc.). Most NER solutions operate in the general or news domain, which makes them inapplicable to the analysis of legal documents. Accordingly, there is a great need for an NER-annotated dataset consisting of legal documents, including the corresponding development of a typology of semantic concepts and uniform annotation guidelines.
Source Data
Court decisions from 2017 and 2018 were selected for the dataset, published online by the Federal Ministry of Justice and Consumer Protection. The documents originate from seven federal courts: Federal Labour Court (BAG), Federal Fiscal Court (BFH), Federal Court of Justice (BGH), Federal Patent Court (BPatG), Federal Social Court (BSG), Federal Constitutional Court (BVerfG) and Federal Administrative Court (BVerwG).
Initial Data Collection and Normalization
From the table of contents, 107 documents from each court were selected (see Table 1). The data was collected from the XML documents, i. e., it was extracted from the XML elements Mitwirkung, Titelzeile, Leitsatz, Tenor, Tatbestand, Entscheidungsgründe, Gründen, abweichende Meinung, and sonstiger Titel. The metadata at the beginning of the documents (name of court, date of decision, file number, European Case Law Identifier, document type, laws) and those that belonged to previous legal proceedings was deleted. Paragraph numbers were removed.
The extracted data was split into sentences, tokenised using SoMaJo and manually annotated in WebAnno.
Who are the source language producers?
The Federal Ministry of Justice and the Federal Office of Justice provide selected decisions. Court decisions were produced by humans.
Annotations
Annotation process
For more details see annotation guidelines (in German).
Personal and Sensitive Information
A fundamental characteristic of the published decisions is that all personal information have been anonymised for privacy reasons. This affects the classes person, location and organization.
Licensing Information
Citation Information
@misc{https://doi.org/10.48550/arxiv.2003.13016,
doi = {10.48550/ARXIV.2003.13016},
url = {https://arxiv.org/abs/2003.13016},
author = {Leitner, Elena and Rehm, Georg and Moreno-Schneider, Julián},
keywords = {Computation and Language (cs.CL), Information Retrieval (cs.IR), FOS: Computer and information sciences, FOS: Computer and information sciences},
title = {A Dataset of German Legal Documents for Named Entity Recognition},
publisher = {arXiv},
year = {2020},
copyright = {arXiv.org perpetual, non-exclusive license}
}
Contributions
- Downloads last month
- 110
Models trained or fine-tuned on elenanereiss/german-ler