Eurlex-Toolbox

This repository contains a python toolbox to load, parse and process Official Journals of the European Union (EU).

EU Law text corpus

⏬ Download the dataset here.

Text corpus containing all the legal acts (the L series of the Official Journal) of the European Union adopted since the entry into force of the Lisbon Treaty, 1st December 2009, until 30th June 2019.

The resulting corpus contains 24134 documents totalling approximately 43 million words.

We present it divided in three files.

• Legislation contains the legal acts the Union can adopt to exercise its competences (those listed under Article 288 Treaty on the Functioning of the European Union): regulations, directives, decisions, recommendations and opinions. It also contains the EU acts implementing them.
• International Agreements and EEA acts contains, in addition to Treaties concluded by the European Union, texts with EEA relevance.
• Other contains guidelines, interinstitutional agreements, notices, procedural rules, etc.

Software Overview

European Union law documents are publicly available in human-readable format in the EUR-Lex portal. To enable automatic analysis, the same documents are released as structured text in the EU Open Data Portal. Options for bulk download of entire blocks of documents are available there.

This software allows to handle the Official Journals of the EU as they are released in XML-Formex format here.

Data preparation

To maximize reproducibility and ease of use, the download and decompression of XML data from here is automatized thanks to the download_and_unzip.py script. Usage:

python download_and_unzip.py <dataset_root> <language>

Once the download_and_unzip script has finished, the expected directory structure should look like the following:

<eurlex_root>/
  ├── JOx_FMX_EN_2004/
    ├── file0.doc.xml
    ├── file0.xml
    ├── ...
    ├── fileN.doc.xml
    └── fileN.xml
  ├── JOx_FMX_EN_2005/
      ├── ...
  ├── ...
  └── JOx_FMX_EN_2019/
      ├── ...

Hello World!

Once the raw XML data are in place, the toolbox (entry point main.py) can be used to manipulate the data and create huamn readable text corpora.

Documents can be dumped in human readable format in few lines of code.

from eurlex_ds import EurLexDataset

dataset = EurLexDataset(data_root=<eurlex_root>)
print(f'Number of documents: {len(dataset)}')

# Dump all concatenated docs in human readable text
dataset.dump_to_txt('all_txt.txt', mode='text')

# Dump all document headers
dataset.dump_to_txt('stats.csv', mode='headers')

NB: To save time, the EurLexDataset object can also be initialized form a text file containing a list of paths pointing to the docs to be loaded (see here). This avoids listing all file on disk every time.

Advanced features

The EurLexDataset object encapsulates the official journals dataset and it can be used to perform more sophisticated analyses and queries. Few examples:

• Create a separate text file for all the documents of each year from 2009 to 2019:

from eurlex_ds import EurLexDataset

dataset = EurLexDataset(data_root=args.data_root)

for year in range(2009, 2020):
  items = [it for it in dataset if it.date.startswith(str(year))]
  year_all_txt = ('\n' * 5).join([it.to_txt() for it in items])
  Path(f'all_txt_{year}.txt').write_text(year_all_txt, encoding='utf-8')

• Filter the dataset keeping only decisions:

from eurlex_ds import EurLexDataset

dataset = EurLexDataset(data_root=args.data_root)
dataset.items[:] = [it for it in dataset if it.meta.is_dec()]

• Print legal value and title of all documents to console:

from eurlex_ds import EurLexDataset

dataset = EurLexDataset(data_root=args.data_root)
for item in dataset:
  print(f'{item.meta.legval}: {item.meta.title}\n')

etc.

Many other features are already available, but still undocumented since we are still working on them and the API might change a lot. These includes tokenization, extraction of geographical entities etc. However, the code is open source, so feel free to explore it and take what you may find useful :)

Citation

The code contained this repository accompanies the following publication:

Palazzi, Andrea and Luigi Lonardo. "A Dataset, Software Toolbox, and Interdisciplinary Research Agenda for the Common Foreign and Security Policy", to appear in the European Foreign Affairs Review Vol. 25, Issue 2 (July 2020).

In case you find this dataset and toolbox helpful for your research (we hope so!), please mention the paper above.