---
license: cc-by-4.0
task_categories:
- token-classification
task_ids:
- named-entity-recognition
language:
- mt
pretty_name: MAPA Maltese
size_categories:
- 1K<n<10K
---

# MAPA Maltese

Named-Entity Recognition dataset from the [MAPA Project](https://mapa-project.eu/).
This dataset has some fixes as detailed in [Cross-Lingual Transfer from Related Languages: Treating Low-Resource Maltese as Multilingual Code-Switching](https://aclanthology.org/2024.eacl-long.61):
- Manually fixed some inconsistencies between Level 1 & Level 2 tags.
- Manually added the labels for some spans which were marked as entity spans but didn't have the tags.
- Manually fixed incorrectly marked spans with respect to tokenisation (either having a sub-word marked as an entity span, or having part of a previous word marked as an entity span; in both cases the whole word should've been marked as a span).
- Re-tokenised the dataset using the [MLRS Tokeniser](https://mlrs.research.um.edu.mt/), mainly done to not split off `-` & `'` characters as separate tokens  as done by the [official convertor](https://gitlab.com/MAPA-EU-Project/mapa_project/-/blob/master/documentation/detection_training.md#converting-inception-tsv-files-to-jsonlines), since these are linguistically important characters in Maltese.
  While doing so, any tokens not split off by the tokeniser but which had multiple entity sub-spans, were also split off into separate tokens.
  Lastly, all tokens ending with `-`/`'` were checked to ensure that these weren't miscellaneous characters (e.g. for number ranges or quotation marks), in which case they were manually split into separate tokens.

For `EurLex` documents the same training/validation/testing splits from [joelniklaus/mapa](https://huggingface.co/datasets/joelniklaus/mapa) are kept.
Otherwise, for the other domains, we split documents in similar ratios.

## Citations

If you used this dataset, please cite these works:
- The original dataset:
  ```bibtex
  @inproceedings{gianola-2020-mapa,
      author = {Lucie Gianola and Ēriks Ajausks and Victoria Arranz and Chomicha Bendahman and Laurent Bié and Claudia Borg and Aleix Cerdà and Khalid Choukri and Montse Cuadros and Ona de Gibert and Hans Degroote and Elena Edelman and  Thierry Etchegoyhen and Ángela Franco Torres and Mercedes García Hernandez and Aitor García Pablos and Albert Gatt and Cyril Grouin and Manuel Herranz and Alejandro Adolfo Kohan and Thomas Lavergne and Maite Melero and Patrick Paroubek and Mickaël Rigault and Mike Rosner and Roberts Rozis and Lonneke van der Plas and Rinalds Vīksna and Pierre Zweigenbaum},
      title = {Automatic Removal of Identifying Information in Official EU Languages for Public Administrations: The {MAPA} Project},
      booktitle = {Proceedings of the 33rd International Conference on Legal Knowledge and Information Systems ({JURIX'20})},
      pages = {223--226},
      year = {2020},
      publisher = {IOS Press},
      url = {https://ebooks.iospress.nl/volumearticle/56182},
      doi = {10.3233/FAIA200869},
  }
  ```
- The fixes & training/validation/testing splits:
  ```bibtex
  @misc{micallef-etal-2024-maltese-etymology,
    title = "Cross-Lingual Transfer from Related Languages: Treating Low-Resource {M}altese as Multilingual Code-Switching",
    author = "Micallef, Kurt  and
              Habash, Nizar  and
              Borg, Claudia  and
              Eryani, Fadhl  and
              Bouamor, Houda",
    editor = "Graham, Yvette  and
              Purver, Matthew",
    booktitle = "Proceedings of the 18th Conference of the European Chapter of the Association for Computational Linguistics (Volume 1: Long Papers)",
    month = mar,
    year = "2024",
    address = "St. Julian{'}s, Malta",
    publisher = "Association for Computational Linguistics",
    url = "https://aclanthology.org/2024.eacl-long.61",
    pages = "1014--1025",
  }
  ```