Dataset Card for CEIL

Dataset Summary

Catalan Entity Identification and Linking (CEIL) is a dataset for complex Named Entity Recognition (NER) created by the AINA project in the BSC for Machine Learning and Language Model evaluation purposes in Catalan. It contains 9 main types and 52 subtypes on all kinds of short texts, with almost 59K documents.

Supported Tasks and Leaderboards

Named Entities Recognition, Language Model

Languages

The dataset is in Catalan (ca-ES).

Dataset Structure

Data Instances

Three two-column files, one for each split.

l'      O
obra    O
de      O
Galileu B-person-scholar/scientist
,       O
i       O
de      O
la      O
multiplicació   O
de      O
les     O
acadèmies       O
científiques    O
,       O
com     O
l'      O
Accademia       B-organization-education
dei     I-organization-education
Lincei  I-organization-education

Data Fields

Every file has two columns, with the word form or punctuation symbol in the first one and the corresponding IOB tag in the second one.

Data Splits

80/20 Train and development sets, balanced for all NERC tags. Test set incloudes documents that contain overall all the possible types in the corpus.

Dataset Creation

Curation Rationale

We created this corpus to contribute to the development of language models in Catalan.

Source Data

Initial Data Collection and Normalization

Documents were gathered from various online sources:

• Tweets about different topics, such as catalan independence, coronavirus, benidormfest, vaccines, etc.
• Newswire from nacio digital (Motor) Vilaweb (opinion pieces), Agencia Catalana de Noticies (Economy and Memoria Histórica)
• Various threads from Raco Catalá forum
• Viquipedia articles (woman bios, film synopses, etc.)
• OTHER: Parliament proceedings, restaurant online reviews, etc.

The word tokenization used to convert offset annotations into CONLL files was done using spacy

Who are the source language producers?

The original data comes from various sources.

Annotations

Annotation process

We adapted the NER labels from to a token-per-line, multi-column format.

Who are the annotators?

The annotation was entrusted to the company M47 labs through a public tender process.

Guidelines available at Zenodo

Personal and Sensitive Information

No personal or sensitive information included.

Considerations for Using the Data

Social Impact of Dataset

We hope this corpus contributes to the development of language models in Catalan, a low-resource language.

Discussion of Biases

[N/A]

Other Known Limitations

[N/A]

Additional Information

Dataset Curators

Language Technologies Unit at the Barcelona Supercomputing Center ([email protected]).

This work has been promoted and financed by the Generalitat de Catalunya through the Aina project.

Licensing Information

CEIL is used under CC-by licence.

Citation Information

@inproceedings{gonzalez-agirre-etal-2024-building-data,
    title = "Building a Data Infrastructure for a Mid-Resource Language: The Case of {C}atalan",
    author = "Gonzalez-Agirre, Aitor  and
      Marimon, Montserrat  and
      Rodriguez-Penagos, Carlos  and
      Aula-Blasco, Javier  and
      Baucells, Irene  and
      Armentano-Oller, Carme  and
      Palomar-Giner, Jorge  and
      Kulebi, Baybars  and
      Villegas, Marta",
    editor = "Calzolari, Nicoletta  and
      Kan, Min-Yen  and
      Hoste, Veronique  and
      Lenci, Alessandro  and
      Sakti, Sakriani  and
      Xue, Nianwen",
    booktitle = "Proceedings of the 2024 Joint International Conference on Computational Linguistics, Language Resources and Evaluation (LREC-COLING 2024)",
    month = may,
    year = "2024",
    address = "Torino, Italia",
    publisher = "ELRA and ICCL",
    url = "https://aclanthology.org/2024.lrec-main.231",
    pages = "2556--2566",
}

Contributions

[N/A]