added first version of mapa dataset

.gitattributes

@@ -39,3 +39,6 @@ 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

README.md

@@ -0,0 +1,273 @@
+---
+annotations_creators:
+- other
+language_creators:
+- found
+languages:
+- bg, cs, da, de, el, en, es, et, fi, fr, ga, hu, it, lt, lv, mt, nl, pt, ro, sk, sv
+license:
+- CC-BY-4.0
+multilinguality:
+- multilingual
+paperswithcode_id: null
+pretty_name: Spanish Datasets for Sensitive Entity Detection in the Legal Domain
+size_categories:
+- 1K<n<10K
+source_datasets:
+- original
+task_categories:
+- token-classification
+task_ids:
+- named-entity-recognition
+- named entity recognition and classification (NERC)
+
+---
+
+# Dataset Card for Spanish Datasets for Sensitive Entity Detection in the Legal Domain
+
+## 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:** [Spanish](https://elrc-share.eu/repository/browse/mapa-anonymization-package-spanish/b550e1a88a8311ec9c1a00155d026706687917f92f64482587c6382175dffd76/), [Most](https://elrc-share.eu/repository/search/?q=mfsp:3222a6048a8811ec9c1a00155d0267067eb521077db54d6684fb14ce8491a391), [German, Portuguese, Slovak, Slovenian, Swedish](https://elrc-share.eu/repository/search/?q=mfsp:833df1248a8811ec9c1a00155d0267067685dcdb77064822b51cc16ab7b81a36)
+- **Paper:** de Gibert Bonet, O., García Pablos, A., Cuadros, M., & Melero, M. (2022). Spanish Datasets for Sensitive
+  Entity Detection in the Legal Domain. Proceedings of the Language Resources and Evaluation Conference, June,
+  3751–3760. http://www.lrec-conf.org/proceedings/lrec2022/pdf/2022.lrec-1.400.pdf
+- **Leaderboard:**
+- **Point of Contact:** [Joel Niklaus](joel.niklaus.2@bfh.ch)
+
+### Dataset Summary
+
+The dataset consists of 12 documents (9 for Spanish due to parsing errors) taken from EUR-Lex, a multilingual corpus of court
+decisions and legal dispositions in the 24 official languages of the European Union. The documents have been annotated
+for named entities following the guidelines of the [MAPA project]( https://mapa-project.eu/) which foresees two
+annotation level, a general and a more fine-grained one. The annotated corpus can be used for named entity recognition/classification.
+
+### Supported Tasks and Leaderboards
+
+The dataset supports the task of Named Entity Recognition and Classification (NERC).
+
+### Languages
+
+The following languages are supported: bg, cs, da, de, el, en, es, et, fi, fr, ga, hu, it, lt, lv, mt, nl, pt, ro, sk, sv
+
+## Dataset Structure
+
+### Data Instances
+
+The file format is jsonl and three data splits are present (train, validation and test). Named Entity annotations are
+non-overlapping.
+
+### Data Fields
+
+For the annotation the documents have been split into sentences. The annotations has been done on the token level. 
+The files contain the following data fields
+
+- `language`: language of the sentence
+- `type`:  The document type of the sentence. Currently, only EUR-LEX is supported.
+- `file_name`: The document file name the sentence belongs to.
+- `sentence_number`: The number of the sentence inside its document.
+- `tokens`: The list of tokens in the sentence.
+- `coarse_grained`: The coarse-grained annotations for each token
+- `fine_grained`: The fine-grained annotations for each token
+
+
+As previously stated, the annotation has been conducted on a global and a more fine-grained level.
+
+The tagset used for the global and the fine-grained named entities is the following:
+
+- Address
+    - Building
+    - City
+    - Country
+    - Place
+    - Postcode
+    - Street
+    - Territory
+- Amount
+    - Unit
+    - Value
+- Date
+    - Year
+    - Standard Abbreviation
+    - Month
+    - Day of the Week
+    - Day
+    - Calender Event
+- Person
+    - Age
+    - Email
+    - Ethnic Category
+    - Family Name
+    - Financial
+    - Given Name – Female
+    - Given Name – Male
+    - Health Insurance Number
+    - ID Document Number
+    - Initial Name
+    - Marital Status
+    - Medical Record Number
+    - Nationality
+    - Profession
+    - Role
+    - Social Security Number
+    - Title
+    - Url
+- Organisation
+- Time
+- Vehicle
+    - Build Year
+    - Colour
+    - License Plate Number
+    - Model
+    - Type
+
+### Data Splits
+
+Splits created by Joel Niklaus.
+
+
+| language   |   # train files |   # validation files |   # test files |   # train sentences |   # validation sentences |   # test sentences |
+|:-----------|----------------:|---------------------:|---------------:|--------------------:|-------------------------:|-------------------:|
+| bg         |               9 |                    1 |              2 |                1411 |                      166 |                560 |
+| cs         |               9 |                    1 |              2 |                1464 |                      176 |                563 |
+| da         |               9 |                    1 |              2 |                1455 |                      164 |                550 |
+| de         |               9 |                    1 |              2 |                1457 |                      166 |                558 |
+| el         |               9 |                    1 |              2 |                1529 |                      174 |                584 |
+| en         |               9 |                    1 |              2 |                 893 |                       98 |                408 |
+| es         |               7 |                    1 |              1 |                 806 |                      248 |                155 |
+| et         |               9 |                    1 |              2 |                1391 |                      163 |                516 |
+| fi         |               9 |                    1 |              2 |                1398 |                      187 |                531 |
+| fr         |               9 |                    1 |              2 |                1297 |                       97 |                490 |
+| ga         |               9 |                    1 |              2 |                1383 |                      165 |                515 |
+| hu         |               9 |                    1 |              2 |                1390 |                      171 |                525 |
+| it         |               9 |                    1 |              2 |                1411 |                      162 |                550 |
+| lt         |               9 |                    1 |              2 |                1413 |                      173 |                548 |
+| lv         |               9 |                    1 |              2 |                1383 |                      167 |                553 |
+| mt         |               9 |                    1 |              2 |                 937 |                       93 |                442 |
+| nl         |               9 |                    1 |              2 |                1391 |                      164 |                530 |
+| pt         |               9 |                    1 |              2 |                1086 |                      105 |                390 |
+| ro         |               9 |                    1 |              2 |                1480 |                      175 |                557 |
+| sk         |               9 |                    1 |              2 |                1395 |                      165 |                526 |
+| sv         |               9 |                    1 |              2 |                1453 |                      175 |                539 |
+
+## Dataset Creation
+
+### Curation Rationale
+
+*„[…] to our knowledge, there exist no open resources annotated for NERC [Named Entity Recognition and Classificatio] in Spanish in the legal domain. With the
+present contribution, we intend to fill this gap. With the release of the created resources for fine-tuning and
+evaluation of sensitive entities detection in the legal domain, we expect to encourage the development of domain-adapted
+anonymisation tools for Spanish in this field“* (de Gibert Bonet et al., 2022)
+
+### Source Data
+
+#### Initial Data Collection and Normalization
+
+The dataset consists of documents taken from EUR-Lex corpus which is publicly available. No further
+information on the data collection process are given in de Gibert Bonet et al. (2022).
+
+#### Who are the source language producers?
+
+The source language producers are presumably lawyers.
+
+### Annotations
+
+#### Annotation process
+
+*"The annotation scheme consists of a complex two level hierarchy adapted to the legal domain, it follows the scheme
+described in (Gianola et al., 2020) […] Level 1 entities refer to general categories (PERSON, DATE, TIME, ADDRESS...)
+and level 2 entities refer to more fine-grained subcategories (given name, personal name, day, year, month...). Eur-Lex,
+CPP and DE have been annotated following this annotation scheme […] The manual annotation was performed using
+INCePTION (Klie et al., 2018) by a sole annotator following the guidelines provided by the MAPA consortium."* (de Gibert
+Bonet et al., 2022)
+
+#### Who are the annotators?
+
+Only one annotator conducted the annotation. More information are not provdided in de Gibert Bonet et al. (2022).
+
+### Personal and Sensitive Information
+
+[More Information Needed]
+
+## Considerations for Using the Data
+
+### Social Impact of Dataset
+
+[More Information Needed]
+
+### Discussion of Biases
+
+[More Information Needed]
+
+### Other Known Limitations
+
+Note that the dataset at hand presents only a small portion of a bigger corpus as described in de Gibert Bonet et al. 
+(2022). At the time of writing only the annotated documents from the EUR-Lex corpus were available.
+
+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
+
+[Attribution 4.0 International (CC BY 4.0) ](https://creativecommons.org/licenses/by/4.0/)
+
+### Citation Information
+
+```
+@article{DeGibertBonet2022,
+author = {{de Gibert Bonet}, Ona and {Garc{\'{i}}a Pablos}, Aitor and Cuadros, Montse and Melero, Maite},
+journal = {Proceedings of the Language Resources and Evaluation Conference},
+number = {June},
+pages = {3751--3760},
+title = {{Spanish Datasets for Sensitive Entity Detection in the Legal Domain}},
+url = {https://aclanthology.org/2022.lrec-1.400},
+year = {2022}
+}
+```
+
+### Contributions
+
+Thanks to [@JoelNiklaus](https://github.com/joelniklaus) and [@kapllan](https://github.com/kapllan) for adding this
+dataset.

convert_to_hf_dataset.py

@@ -0,0 +1,190 @@
+import os
+from glob import glob
+from pathlib import Path
+
+import numpy as np
+import pandas as pd
+
+from web_anno_tsv import open_web_anno_tsv
+from web_anno_tsv.web_anno_tsv import ReadException, Annotation
+
+pd.set_option('display.max_colwidth', None)
+pd.set_option('display.max_columns', None)
+
+annotation_labels = {'ADDRESS': ['building', 'city', 'country', 'place', 'postcode', 'street', 'territory'],
+                     'AMOUNT': ['unit', 'value'],
+                     'DATE': ['year', 'standard abbreviation', 'month', 'day of the week', 'day', 'calender event'],
+                     'PERSON': ['age', 'email', 'ethnic category', 'family name', 'financial', 'given name – female',
+                                'given name – male',
+                                'health insurance number', 'id document number', 'initial name', 'marital status',
+                                'medical record number',
+                                'nationality', 'profession', 'role', 'social security number', 'title', 'url'],
+                     'ORGANISATION': [],
+                     'TIME': [],
+                     'VEHICLE': ['build year', 'colour', 'license plate number', 'model', 'type']}
+
+# make all coarse_grained upper case and all fine_grained lower case
+annotation_labels = {key.upper(): [label.lower() for label in labels] for key, labels in annotation_labels.items()}
+print(annotation_labels)
+
+base_path = Path("extracted")
+
+# TODO future work can add these datasets too to make it larger
+special_paths = {
+    "EL": ["EL/ANNOTATED_DATA/LEGAL/AREIOSPAGOS1/annotated/full_dataset"],
+    "EN": ["EN/ANNOTATED_DATA/ADMINISTRATIVE-LEGAL/annotated/full_dataset"],
+    "FR": ["FR/ANNOTATED_DATA/LEGAL/COUR_CASSATION1/annotated/full_dataset/Civil",
+           "FR/ANNOTATED_DATA/LEGAL/COUR_CASSATION1/annotated/full_dataset/Commercial",
+           "FR/ANNOTATED_DATA/LEGAL/COUR_CASSATION1/annotated/full_dataset/Criminal",
+           "FR/ANNOTATED_DATA/LEGAL/COUR_CASSATION2/annotated/full_dataset",
+           "FR/ANNOTATED_DATA/MEDICAL/CAS1/annotated/full_dataset"],
+    "IT": ["IT/ANNOTATED_DATA/Corte_Suprema_di_Cassazione/annotated"],
+    "MT": ["MT/ANNOTATED_DATA/ADMINISTRATIVE/annotated/full_dataset",
+           "MT/ANNOTATED_DATA/GENERAL_NEWS/News_1/annotated/full_dataset",
+           "MT/ANNOTATED_DATA/LEGAL/Jurisprudence_1/annotated/full_dataset"],
+}
+
+
+def get_path(language):
+    return base_path / language / "ANNOTATED_DATA/EUR_LEX/annotated/full_dataset"
+
+
+def get_coarse_grained_for_fine_grained(label):
+    for coarse_grained, fine_grained_set in annotation_labels.items():
+        if label in fine_grained_set:
+            return coarse_grained
+    return None  # raise ValueError(f"Did not find fine_grained label {label}")
+
+
+def is_fine_grained(label):
+    for coarse_grained, fine_grained_set in annotation_labels.items():
+        if label.lower() in fine_grained_set:
+            return True
+    return False
+
+
+def is_coarse_grained(label):
+    return label.upper() in annotation_labels.keys()
+
+
+class HashableAnnotation(Annotation):
+    def __init__(self, annotation):
+        super()
+        self.label = annotation.label
+        self.start = annotation.start
+        self.stop = annotation.stop
+        self.text = annotation.text
+
+    def __eq__(self, other):
+        return self.label == other.label and self.start == other.start and self.stop == other.stop and self.text == other.text
+
+    def __hash__(self):
+        return hash(('label', self.label, 'start', self.start, 'stop', self.stop, 'text', self.text))
+
+
+def get_token_annotations(token, annotations):
+    annotations = list(dict.fromkeys([HashableAnnotation(ann) for ann in annotations]))  # remove duplicate annotations
+    coarse_grained = "O"
+    fine_grained = "o"
+    for annotation in annotations:
+        label = annotation.label
+        # if token.start == annotation.start and token.stop == annotation.stop:  # fine_grained annotation
+        if token.start >= annotation.start and token.stop <= annotation.stop:  # course_grained annotation
+            # we don't support multilabel annotations for each token for simplicity.
+            # So when a token already has an annotation for either coarse or fine grained, we don't assign new ones.
+            if coarse_grained != "O" and is_coarse_grained(label):
+                coarse_grained = label
+            elif fine_grained != "o" and is_fine_grained(label):
+                # some DATE are mislabeled as day but it is hard to correct this. So we ignore it
+                fine_grained = label
+
+    return coarse_grained.upper(), fine_grained.lower()
+
+
+def get_annotated_sentence(result_sentence, sentence):
+    result_sentence["tokens"] = []
+    result_sentence["coarse_grained"] = []
+    result_sentence["fine_grained"] = []
+    for k, token in enumerate(sentence.tokens):
+        coarse_grained, fine_grained = get_token_annotations(token, sentence.annotations)
+        token = token.text.replace(u'\xa0', u' ').strip()  # replace non-breaking spaces
+        if token:  # remove empty tokens (only consisted of whitespace before
+            result_sentence["tokens"].append(token)
+            result_sentence["coarse_grained"].append(coarse_grained)
+            result_sentence["fine_grained"].append(fine_grained)
+    return result_sentence
+
+
+languages = sorted([Path(file).stem for file in glob(str(base_path / "*"))])
+
+
+def parse_files(language):
+    data_path = get_path(language.upper())
+    result_sentences = []
+    not_parsable_files = 0
+    file_names = sorted(list(glob(str(data_path / "*.tsv"))))
+    for file in file_names:
+        try:
+            with open_web_anno_tsv(file) as f:
+                for i, sentence in enumerate(f):
+                    result_sentence = {"language": language, "type": "EUR-LEX",
+                                       "file_name": Path(file).stem, "sentence_number": i}
+                    result_sentence = get_annotated_sentence(result_sentence, sentence)
+                    result_sentences.append(result_sentence)
+            print(f"Successfully parsed file {file}")
+        except ReadException as e:
+            print(f"Could not parse file {file}")
+            not_parsable_files += 1
+    print("Not parsable files: ", not_parsable_files)
+    return pd.DataFrame(result_sentences), not_parsable_files
+
+
+stats = []
+train_dfs, validation_dfs, test_dfs = [], [], []
+for language in languages:
+    language = language.lower()
+    print(f"Parsing language {language}")
+    df, not_parsable_files = parse_files(language)
+    file_names = df.file_name.unique()
+
+    # split by file_name
+    num_fn = len(file_names)
+    train_fn, validation_fn, test_fn = np.split(np.array(file_names), [int(.8 * num_fn), int(.9 * num_fn)])
+
+    lang_train = df[df.file_name.isin(train_fn)]
+    lang_validation = df[df.file_name.isin(validation_fn)]
+    lang_test = df[df.file_name.isin(test_fn)]
+
+    train_dfs.append(lang_train)
+    validation_dfs.append(lang_validation)
+    test_dfs.append(lang_test)
+
+    lang_stats = {"language": language}
+
+    lang_stats["# train files"] = len(train_fn)
+    lang_stats["# validation files"] = len(validation_fn)
+    lang_stats["# test files"] = len(test_fn)
+
+    lang_stats["# train sentences"] = len(lang_train.index)
+    lang_stats["# validation sentences"] = len(lang_validation.index)
+    lang_stats["# test sentences"] = len(lang_test.index)
+
+    stats.append(lang_stats)
+
+stat_df = pd.DataFrame(stats)
+print(stat_df.to_markdown(index=False))
+
+train = pd.concat(train_dfs)
+validation = pd.concat(validation_dfs)
+test = pd.concat(test_dfs)
+
+# save splits
+def save_splits_to_jsonl(config_name):
+    # save to jsonl files for huggingface
+    if config_name: os.makedirs(config_name, exist_ok=True)
+    train.to_json(os.path.join(config_name, "train.jsonl"), lines=True, orient="records", force_ascii=False)
+    validation.to_json(os.path.join(config_name, "validation.jsonl"), lines=True, orient="records", force_ascii=False)
+    test.to_json(os.path.join(config_name, "test.jsonl"), lines=True, orient="records", force_ascii=False)
+
+
+save_splits_to_jsonl("")

test.jsonl

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:55e08959d49e88b52f29395b7515e8c456eddbd84592b0ea71a49326c322348f
+size 7559023

train.jsonl

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:9ea732c32b177da86de291522d0a4236ff53f92939ad7b2762d3d8d8e4449ba6
+size 21505824

validation.jsonl

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:1e8d04f631b1e15806bdc54fd01eabc35017c4645bfd524ccadfb980937b592d
+size 2797014