added dataset files

.gitattributes

@@ -35,3 +35,7 @@ saved_model/**/* filter=lfs diff=lfs merge=lfs -text
 *.mp3 filter=lfs diff=lfs merge=lfs -text
 *.ogg filter=lfs diff=lfs merge=lfs -text
 *.wav filter=lfs diff=lfs merge=lfs -text
+test.jsonl filter=lfs diff=lfs merge=lfs -text
+train.jsonl filter=lfs diff=lfs merge=lfs -text
+validation.jsonl filter=lfs diff=lfs merge=lfs -text
+meta.jsonl filter=lfs diff=lfs merge=lfs -text

README.md

@@ -0,0 +1,260 @@
+---
+annotations_creators:
+- expert-generated
+- found
+language_creators:
+- found
+languages:
+- de
+license:
+- cc-by-4.0
+multilinguality:
+- monolingual
+pretty_name: Annotated German Legal Decision Corpus
+size_categories:
+- 10K<n<100K
+source_datasets:
+- original
+task_categories:
+- text-classification
+task_ids:
+- multi-class-classification
+---
+
+# Dataset Card for Annotated German Legal Decision Corpus
+
+## Table of Contents
+
+- [Table of Contents](#table-of-contents)
+- [Dataset Description](#dataset-description)
+  - [Dataset Summary](#dataset-summary)
+  - [Supported Tasks and Leaderboards](#supported-tasks-and-leaderboards)
+  - [Languages](#languages)
+- [Dataset Structure](#dataset-structure)
+  - [Data Instances](#data-instances)
+  - [Data Fields](#data-fields)
+  - [Data Splits](#data-splits)
+- [Dataset Creation](#dataset-creation)
+  - [Curation Rationale](#curation-rationale)
+  - [Source Data](#source-data)
+  - [Annotations](#annotations)
+  - [Personal and Sensitive Information](#personal-and-sensitive-information)
+- [Considerations for Using the Data](#considerations-for-using-the-data)
+  - [Social Impact of Dataset](#social-impact-of-dataset)
+  - [Discussion of Biases](#discussion-of-biases)
+  - [Other Known Limitations](#other-known-limitations)
+- [Additional Information](#additional-information)
+  - [Dataset Curators](#dataset-curators)
+  - [Licensing Information](#licensing-information)
+  - [Citation Information](#citation-information)
+  - [Contributions](#contributions)
+
+## Dataset Description
+
+- **Homepage:**
+- **Repository:** https://zenodo.org/record/3936490#.X1ed7ovgomK
+- **Paper:** Urchs., S., Mitrović., J., & Granitzer., M. (2021). Design and Implementation of German Legal Decision
+  Corpora. Proceedings of the 13th International Conference on Agents and Artificial Intelligence - Volume 2: ICAART,
+  515–521. https://doi.org/10.5220/0010187305150521
+- **Leaderboard:**
+- **Point of Contact:** [Joel Niklaus](joel.niklaus.2@bfh.ch)
+
+### Dataset Summary
+
+This dataset consists of 200 randomly chosen judgments. In these judgments a legal expert annotated the components
+conclusion, definition and subsumption of the German legal writing style Urteilsstil.
+
+*"Overall 25,075 sentences are annotated. 5% (1,202) of these sentences are marked as conclusion, 21% (5,328) as
+definition, 53% (13,322) are marked as subsumption and the remaining 21% (6,481) as other. The length of judgments in
+sentences ranges from 38 to 862 sentences. The median of judgments have 97 sentences, the length of most judgments is on
+the shorter side."* (Urchs. et al., 2021)
+
+*"Judgments from 22 of the 131 courts are selected for the corpus. Most judgments originate from the VG Augsburg (59 /
+30%) followed by the VG Ansbach (39 / 20%) and LSG Munich (33 / 17%)."* (Urchs. et al., 2021)
+
+*"29% (58) of all selected judgments are issued in the year 2016, followed by 22% (44) from the year 2017 and 21% (41)
+issued in the year 2015. [...] The percentages of selected judgments and decisions issued in 2018 and 2019 are roughly
+the same. No judgments from 2020 are selected."* (Urchs. et al., 2021)
+
+### Supported Tasks and Leaderboards
+
+The dataset can be used for multi-class text classification tasks, more specifically, for argument mining.
+
+### Languages
+
+The language in the dataset is German as it is used in Bavarian courts in Germany.
+
+## Dataset Structure
+
+### Data Instances
+
+Each sentence is saved as a json object on a line in one of the three files `train.jsonl`, `validation.jsonl`
+or `test.jsonl`. The file `meta.jsonl` contains meta information for each court. The `file_number` is present in all
+files for identification. Each sentence of the court decision was categorized according to its function.
+
+### Data Fields
+
+The file `meta.jsonl` contains for each row the following fields:
+
+- `meta_title`: Title provided by the website, it is used for saving the decision
+- `court`: Issuing court
+- `decision_style`: Style of the decision; the corpus contains either *Urteil* (='judgment') or *Endurteil* (
+  ='end-judgment')
+- `date`: Date when the decision was issued by the court
+- `file_number`: Identification number used for this decision by the court
+- `title`: Title provided by the court
+- `norm_chains`: Norms related to the decision
+- `decision_guidelines`: Short summary of the decision
+- `keywords`: Keywords associated with the decision
+- `lower_court`: Court that decided on the decision before
+- `additional_information`: Additional Information
+- `decision_reference`: References to the location of the decision in beck-online
+- `tenor`: Designation of the legal consequence ordered by the court (list of paragraphs)
+- `legal_facts`: Facts that form the base for the decision (list of paragraphs)
+
+The files `train.jsonl`, `validation.jsonl` and `test.jsonl` contain the following fields:
+
+- `file_number`: Identification number for linkage with the file `meta.jsonl`
+- `input_sentence`: The sentence to be classified
+- `label`: In depth explanation of the court decision. Each sentence is assigned to one of the major components of
+  German *Urteilsstil* (Urchs. et al., 2021) (list of paragraphs, each paragraph containing list of sentences, each
+  sentence annotated with one of the following four labels):
+  - `conclusion`: Overall result
+  - `definition`: Abstract legal facts and consequences
+  - `subsumption`: Determination sentence / Concrete facts
+  - `other`: Anything else
+- `context_before`: Context in the same paragraph before the input_sentence
+- `context_after`: Context in the same paragraph after the input_sentence
+
+### Data Splits
+
+No split provided in the original release.
+
+Splits created by Joel Niklaus. We randomly split the dataset into 80% (160 decisions, 19271 sentences) train, 10%
+validation (20 decisions, 2726 sentences) and 10% test (20 decisions, 3078 sentences). We made sure, that a decision
+only occurs in one split and is not dispersed over multiple splits.
+
+Label Distribution
+
+| label          |      train |   validation |      test |
+|:---------------|-----------:|-------------:|----------:|
+| conclusion     |        975 |          115 |       112 |
+| definition     |       4105 |          614 |       609 |
+| subsumption    |      10034 |         1486 |      1802 |
+| other          |       4157 |          511 |       555 |
+| total          |  **19271** |     **2726** |  **3078** |
+
+## Dataset Creation
+
+### Curation Rationale
+
+Creating a publicly available German legal text corpus consisting of judgments that have been annotated by a legal
+expert. The annotated components consist of *conclusion*, *definition* and *subsumption* of the German legal writing
+style *Urteilsstil*.
+
+### Source Data
+
+#### Initial Data Collection and Normalization
+
+*“The decision corpus is a collection of the decisions published on the website www.gesetze-bayern.de. At the time of
+the crawling the website offered 32,748 decisions of 131 Bavarian courts, dating back to 2015. The decisions are
+provided from the Bavarian state after the courts agreed to a publication. All decisions are processed by the publisher
+C.H.BECK, commissioned by the Bavarian state. This processing includes anonymisation, key-wording, and adding of
+editorial guidelines to the decisions.”* (Urchs. et al., 2021)
+
+#### Who are the source language producers?
+
+German courts from Bavaria
+
+### Annotations
+
+#### Annotation process
+
+*“As stated above, the judgment corpus consist of 200 randomly chosen judgments that are annotated by a legal expert,
+who holds a first legal state exam. Due to financial, staff and time reasons the presented iteration of the corpus was
+only annotated by a single expert. In a future version several other experts will annotate the corpus and the
+inter-annotator agreement will be calculated.”* (Urchs. et al., 2021)
+
+#### Who are the annotators?
+
+A legal expert, who holds a first legal state exam.
+
+### Personal and Sensitive Information
+
+*"All decisions are processed by the publisher C.H.BECK, commissioned by the Bavarian state. This processing includes **
+anonymisation**, key-wording, and adding of editorial guidelines to the decisions.”* (Urchs. et al., 2021)
+
+## Considerations for Using the Data
+
+### Social Impact of Dataset
+
+[More Information Needed]
+
+### Discussion of Biases
+
+[More Information Needed]
+
+### Other Known Limitations
+
+The SoMaJo Sentence Splitter has been used. Upon manual inspection of the dataset, we could see that the sentence
+splitter had poor accuracy in some cases (see ```analyze_dataset()``` in ```convert_to_hf_dataset.py```). When creating
+the splits, we thought about merging small sentences with their neighbors or removing them all together. However, since
+we could not find an straightforward way to do this, we decided to leave the dataset content untouched.
+
+Note that the information given in this dataset card refer to the dataset version as provided by Joel Niklaus and Veton
+Matoshi. The dataset at hand is intended to be part of a bigger benchmark dataset. Creating a benchmark dataset
+consisting of several other datasets from different sources requires postprocessing. Therefore, the structure of the
+dataset at hand, including the folder structure, may differ considerably from the original dataset. In addition to that,
+differences with regard to dataset statistics as give in the respective papers can be expected. The reader is advised to
+have a look at the conversion script ```convert_to_hf_dataset.py``` in order to retrace the steps for converting the
+original dataset into the present jsonl-format. For further information on the original dataset structure, we refer to
+the bibliographical references and the original Github repositories and/or web pages provided in this dataset card.
+
+## Additional Information
+
+### Dataset Curators
+
+The names of the original dataset curators and creators can be found in references given below, in the section *Citation
+Information*. Additional changes were made by Joel Niklaus ([Email](joel.niklaus.2@bfh.ch)
+; [Github](https://github.com/joelniklaus)) and Veton Matoshi ([Email](veton.matoshi@bfh.ch)
+; [Github](https://github.com/kapllan)).
+
+### Licensing Information
+
+[Creative Commons Attribution 4.0 International](https://creativecommons.org/licenses/by/4.0/legalcode)
+
+### Citation Information
+
+```
+@dataset{urchs_stefanie_2020_3936490,
+  author       = {Urchs, Stefanie and
+                  Mitrović, Jelena},
+  title        = {{German legal jugements annotated with judement 
+                   style components}},
+  month        = jul,
+  year         = 2020,
+  publisher    = {Zenodo},
+  doi          = {10.5281/zenodo.3936490},
+  url          = {https://doi.org/10.5281/zenodo.3936490}
+}
+```
+
+```
+@conference{icaart21,
+  author = {Urchs., Stefanie and Mitrovi{\'{c}}., Jelena and Granitzer., Michael},
+  booktitle = {Proceedings of the 13th International Conference on Agents and Artificial Intelligence - Volume 2: ICAART,},
+  doi = {10.5220/0010187305150521},
+  isbn = {978-989-758-484-8},
+  issn = {2184-433X},
+  organization = {INSTICC},
+  pages = {515--521},
+  publisher = {SciTePress},
+  title = {{Design and Implementation of German Legal Decision Corpora}},
+  year = {2021}
+}
+```
+
+### Contributions
+
+Thanks to [@kapllan](https://github.com/kapllan) and [@joelniklaus](https://github.com/joelniklaus) for adding this
+dataset.

convert_to_hf_dataset.py

@@ -0,0 +1,138 @@
+from glob import glob
+from pathlib import Path
+
+import json
+import numpy as np
+import pandas as pd
+
+"""
+Dataset url: https://zenodo.org/record/3936490/files/annotated_corpus.zip?download=1
+Paper url: https://www.ncbi.nlm.nih.gov/pmc/articles/PMC9044329/
+
+There are no splits available ==> Make random split ourselves
+
+"""
+
+pd.set_option('display.max_colwidth', None)
+pd.set_option('display.max_columns', None)
+
+
+def analyze_dataset(df, num_characters_for_short_sentence=25):
+    short_sentence = False
+    counter = 0
+    same_label_counter = 0
+    other_number = 0
+    num_one_paragraph_len = 0
+    for i in range(200):
+        for paragraph in df.iloc[i].decision_reasons:
+            for sentence in paragraph:
+                if short_sentence:
+                    print("previous sentence was short: ", short_sentence)
+                    print("current sentence label: ", sentence[1])
+                    print("current paragraph: ", paragraph)
+                    if sentence[1] == short_sentence[1]:
+                        same_label_counter += 1
+
+                if len(sentence[0]) < num_characters_for_short_sentence:
+                    counter += 1
+                    short_sentence = sentence
+                    print()
+                    print("short sentence: ", sentence)
+                    print("short paragraph: ", paragraph)
+                    if sentence[1] == 'other':
+                        other_number += 1
+                    if len(paragraph) == 1:
+                        num_one_paragraph_len += 1
+                else:
+                    short_sentence = False
+
+    print("num short sentences: ", counter)
+    print("num short sentences containing the same label as the next one: ", same_label_counter)
+    print("num short sentences containing 'other' as label: ", other_number)
+    print("num short sentences where the paragraph contains only this one short sentence: ", num_one_paragraph_len)
+    # ==> the label is only the same in very few cases
+    # ==> the label is 'other' in the majority of cases, when it is not: it seems to be mislabeled
+    # ==> think about removing them entirely
+    # ==> we opted for not interfering in the content of the dataset
+
+
+# create a summary jsonl file
+dataset_filename = "dataset.jsonl"
+if not Path(dataset_filename).exists():
+    with open(dataset_filename, "a") as dataset_file:
+        for filename in glob("annotated_corpus/*.json"):
+            # we need to do this charade, because some jsons are formatted differently than others
+            json_text = Path(filename).read_text()
+            json_obj = json.loads(json_text)
+            # make it less nested so that it is easier to read as df
+            new_dict = {}
+            new_dict.update(json_obj["meta"])
+            new_dict.update(json_obj["decision_text"])
+            dataset_file.write(json.dumps(new_dict) + "\n")
+else:
+    print(f"{dataset_filename} already exists. Please delete it to re-aggregate it.")
+
+df = pd.read_json(dataset_filename, lines=True)
+
+# Do splits before expanding the df so that entire decisions are in the splits and not samples from one decision are spread across splits
+# perform random split 80% train (160 decisions), 10% validation (20 decisions), 10% test (20 decisions)
+train, validation, test = np.split(df.sample(frac=1, random_state=42), [int(.8 * len(df)), int(.9 * len(df))])
+
+
+def expand_df(df):
+    """
+    Expand the df so that each sentence has its own row and is its own sample
+    :param df:
+    :return:
+    """
+    rows = []
+    for index, row in df.iterrows():
+        for paragraph in row.decision_reasons:
+            for sent_idx, sentence in enumerate(paragraph):
+                new_row = {'file_number': row['file_number'], 'input_sentence': sentence[0], 'label': sentence[1]}
+                # Discussion with lawyer yielded, that the paragraph as context is enough
+                # take the sentences before
+                new_row['context_before'] = paragraph[:sent_idx]
+                # take the remaining sentences afterwards
+                new_row['context_after'] = paragraph[sent_idx + 1:]
+                rows.append(new_row)
+
+    return pd.DataFrame.from_records(rows)
+
+
+train = expand_df(train)
+validation = expand_df(validation)
+test = expand_df(test)
+
+# Num samples for each split: train (19271), validation (2726), test (3078)
+print(len(train.index), len(validation.index), len(test.index))
+
+# save to jsonl files for huggingface
+train.to_json("train.jsonl", lines=True, orient="records")
+validation.to_json("validation.jsonl", lines=True, orient="records")
+test.to_json("test.jsonl", lines=True, orient="records")
+
+# save main df with meta information to file
+# link to splits is given via file_number
+df = df.drop(['decision_reasons'], axis=1)
+df.to_json("meta.jsonl", lines=True, orient="records")
+
+
+def print_split_table_single_label(train, validation, test, label_name):
+    train_counts = train[label_name].value_counts().to_frame().rename(columns={label_name: "train"})
+    validation_counts = validation[label_name].value_counts().to_frame().rename(columns={label_name: "validation"})
+    test_counts = test[label_name].value_counts().to_frame().rename(columns={label_name: "test"})
+
+    table = train_counts.join(validation_counts)
+    table = table.join(test_counts)
+    table[label_name] = table.index
+    total_row = {label_name: "total",
+                 "train": len(train.index),
+                 "validation": len(validation.index),
+                 "test": len(test.index)}
+    table = table.append(total_row, ignore_index=True)
+    table = table[[label_name, "train", "validation", "test"]]  # reorder columns
+    print(table.to_markdown(index=False))
+
+
+print_split_table_single_label(train, validation, test, "label")

meta.jsonl

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:83f23fed08311a81117bb12f60be1676313f7448ef297ef5fd34cf4641345e82
+size 2828194

test.jsonl

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:19595a310f39fd1daa89f834930923ed19e3b01d04806d189c382eca70ecbb08
+size 4263762

train.jsonl

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:ed22caee3516ab681121b5896e74de3aceb771d3148587675cb04d4a9d4162d0
+size 26669050

validation.jsonl

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:cf7c5165d3c7d7a42d5d10be6a35391ffd72781dd4008bc6fa15a9daa983d850
+size 3563814