README.md

---
dataset_info:
  features:
  - name: celex_id
    dtype: string
  - name: lang
    dtype: string
  - name: input_text
    dtype: string
  - name: keyphrases
    sequence: string
  splits:
  - name: train
    num_bytes: 4037797726
    num_examples: 131076
  - name: valid
    num_bytes: 2622393019
    num_examples: 63373
  - name: test
    num_bytes: 4781705320
    num_examples: 90508
  download_size: 5173271419
  dataset_size: 11441896065
configs:
- config_name: default
  data_files:
  - split: train
    path: data/train-*
  - split: valid
    path: data/valid-*
  - split: test
    path: data/test-*
license: mit
language:
- fr
- de
- en
- it
- nl
- el
- da
- pt
- es
- sv
- fi
- lt
- et
- cs
- hu
- lv
- sl
- pl
- mt
- sk
- ro
- bg
- hr
- ga
tags:
- keyphrase-generation
- text-to-text
- legal
pretty_name: Europa
size_categories:
- 100K<n<1M
---

# Dataset Card for EUROPA

This dataset card aims to be a base template for new datasets. It has been generated using [this raw template](https://github.com/huggingface/huggingface_hub/blob/main/src/huggingface_hub/templates/datasetcard_template.md?plain=1).

## Dataset Details

### Dataset Description

EUROPA is a dataset designed for training and evaluating multilingual keyphrase generation models in the legal domain. It consists of legal judgments from the Court of Justice of the European Union (EU) and includes instances in all 24 official EU languages.

**Key Features**:
**Multilingual:** Covers 24 official EU languages.
**Domain-Specific:** Focuses on legal documents.
**Source:** Derived from Court of Justice of the European Union judgments.

- **Curated by:** N3 team
- **Languages:** French, German, English, Italian, Dutch, Greek, Danish, Portuguese, Spanish, Swedish, Finnish, Lithuanian, Estonian, Czech, Hungarian, Latvian, Slovenian, Polish, Maltese, Slovak, Romanian, Bulgarian, Croatian, Irish
- **License:** MIT License

### Dataset Sources

- **Paper:** https://aclanthology.org/2024.acl-long.687/

## Dataset Structure

- **celex_id:** CELEX identifier inherited from CJEU. Different translated versions of the same judgment share the same celex_id. If you wish to set a unique identifier for each instance, you can concatenate `lang` and `celex_id` values;
- **lang:** ISO 639-1 language code;
- **input:** judgment transcription or translation;
- **keyphrases:** reference keyphrases drafted by the CJEU.

As explained in our paper, the dataset is split chronologically for assessing temporal generalization of models:
- **training set**: 1957 to 2010 (131 076 instances);
- **validation set**: 2011 to 2015 (63 373 instances);
- **test set**: 2016 to 2023 (90 508 instances).

## Citation

```
@inproceedings{salaun-etal-2024-europa,
    title = "{EUROPA}: A Legal Multilingual Keyphrase Generation Dataset",
    author = {Sala{\"u}n, Olivier  and
      Piedboeuf, Fr{\'e}d{\'e}ric  and
      Le Berre, Guillaume  and
      Alfonso-Hermelo, David  and
      Langlais, Philippe},
    editor = "Ku, Lun-Wei  and
      Martins, Andre  and
      Srikumar, Vivek",
    booktitle = "Proceedings of the 62nd Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers)",
    month = aug,
    year = "2024",
    address = "Bangkok, Thailand",
    publisher = "Association for Computational Linguistics",
    url = "https://aclanthology.org/2024.acl-long.687",
    pages = "12718--12736",
    abstract = "Keyphrase generation has primarily been explored within the context of academic research articles, with a particular focus on scientific domains and the English language. In this work, we present EUROPA, a novel dataset for multilingual keyphrase generation in the legal domain. It is derived from legal judgments from the Court of Justice of the European Union (EU), and contains instances in all 24 EU official languages. We run multilingual models on our corpus and analyze the results, showing room for improvement on a domain-specific multilingual corpus such as the one we present.",
}
```