Dataset Card for "x-fact"

Table of Contents

• Dataset Description
  • Dataset Summary
  • Languages
  • Data Splits
• Additional Information
  • Citation Information

Dataset Description

Dataset Summary

X-FACT is a multilingual dataset for fact-checking with real world claims. The dataset contains short statments in 25 languages with top five evidence documents retrieved by performing google search with claim statements. The dataset contains two additional evaluation splits (in addition to a traditional test set): ood and zeroshot. ood measures out-of-domain generalization where while the language is present in the training data, the claim statements are taken from a different fact-checker. zeroshot on the other hand measures the cross lingual abilities to transfer to new languages. In total, the dataset contains 31,189 claim statments.

Github: https://github.com/utahnlp/x-fact

Languages

The dataset contains claims from 25 languages (with and without training data):

With training data: ar, de, es, hi, id, it, ka, pl, pt, ro, sr, ta, tr

Without training data (only zeroshot): az, bn, fa, fr, gu, mr, nl, no, pa, ru, si, sq

Data Splits

Additional Information

Citation Information

@inproceedings{gupta2021xfact,
      title={{X-FACT: A New Benchmark Dataset for Multilingual Fact Checking}}, 
      author={Gupta, Ashim and Srikumar, Vivek},
      booktitle = "Proceedings of the 59th Annual Meeting of the Association for Computational Linguistics",      
      month = jul,
      year = "2021",
      address = "Online",
      publisher = "Association for Computational Linguistics",
}