RelEx-PT

language:
  - pt
configs:
  - config_name: default
    data_files:
      - split: train
        path: train.jsonl
      - split: test
        path: test.jsonl

RelEx-PT

Dataset Summary

RelEx-PT, a new sentence-level Relation Extraction dataset for Portuguese. Addressing the scarcity of high-quality, controlled resources for the language, RelEx-PT provides a balanced benchmark comprising 18 Wikidata-derived relation types across diverse domains. The dataset is built through a distant supervision pipeline that links Wikidata triples with Portuguese Wikipedia sentences and enhanced by a Natural Language Inference (NLI)-based filtering process, combining scalability with quality assurance.

Dataset Creation

For full construction pipeline and source code check: https://github.com/TomasCCPinto/RelEx-PT

Licensing Information

RelEx-PT is released under the Creative Commons Attribution-ShareAlike 4.0 International (CC BY-SA 4.0) license.

Citation

If you use RelEx-PT in your research, please cite:

@inproceedings{pinto_etal:lrec2026,
    author = {Tom{\´a}s Pinto and Catarina Silva and Hugo {Gon{\c c}alo Oliveira}},
    booktitle = {Proceedings of 15th Language Resources and Evaluation Conference},
    pages = {Accepted},
    publisher = {ELRA},
    series = {LREC 2026},
    title = {{RelEx-PT}: {A} {P}ortuguese {S}entence-{L}evel {R}elation {E}xtraction {D}ataset},
    year = {2026}}