HiTZ/xnli-eu

Dataset Card for XNLIeu

XNLIeu is an extension of XNLI translated from English to Basque. It has been designed as a cross-lingual dataset for the Natural Language Inference task, a text-classification task that consists on classifying pairs of sentences, a premise and a hypothesis, according to their semantic relation out of three possible labels: entailment, contradiction and neutral.

Dataset Details

Dataset Description

XNLI is a popular Natural Language Inference (NLI) benchmark widely used to evaluate cross-lingual Natural Language Understanding (NLU) capabilities across languages. We expand XNLI to include Basque, a low-resource language that can greatly benefit from transfer-learning approaches. The new dataset, dubbed XNLIeu, has been developed by first machine-translating the English XNLI corpus into Basque, followed by a manual post-edition step.

• Language(s) (NLP): Basque (eu)
• License: XNLIeu is derived from XNLI and distributed under its same license.

Dataset Sources

• Repository: Link to the GitHub Repository
• Paper: Link to the Paper

Uses

XNLieu is meant as an cross-lingual evaluation dataset. It can be used in combination with the train sets of XNLI for a cross-lingual zero-shot setting, and we provide a machine-translated train set in both "eu" and "eu_mt" splits to implement a translate-train setting.

Dataset Structure

The dataset has three subsets:

• eu: XNLIeu, machine-translated and post-edited from English to Basque.
• eu_MT: XNLIeu_MT, a machine-translated version prior post-edition.
• eu_native: An original, non-translated test set.

Splits

name	train	validation	test
eu	392702	2490	5010
eu_mt	392702	2490	5010
eu_native	-	-	621

Dataset Fields

All splits have the same fields: premise, hypothesis and label.

• premise: a string variable.
• hypothesis: a string variable.
• label: a classification label, with possible values including entailment (0), neutral (1), contradiction (2).

Dataset Instances

An example from the "eu" split:

{
    "premise": "Dena idazten saiatu nintzen"
    "hypothesis": "Nire helburua gauzak idaztea zen.",
    "label": 0,
}

Bias, Risks, and Limitations

The biases of this dataset have been studied and reported in the paper.

BibTeX:

@inproceedings{heredia-etal-2024-xnlieu,
    title = "{XNLI}eu: a dataset for cross-lingual {NLI} in {B}asque",
    author = "Heredia, Maite  and
      Etxaniz, Julen  and
      Zulaika, Muitze  and
      Saralegi, Xabier  and
      Barnes, Jeremy  and
      Soroa, Aitor",
    editor = "Duh, Kevin  and
      Gomez, Helena  and
      Bethard, Steven",
    booktitle = "Proceedings of the 2024 Conference of the North American Chapter of the Association for Computational Linguistics: Human Language Technologies (Volume 1: Long Papers)",
    month = jun,
    year = "2024",
    address = "Mexico City, Mexico",
    publisher = "Association for Computational Linguistics",
    url = "https://aclanthology.org/2024.naacl-long.234",
    pages = "4177--4188",
    abstract = "XNLI is a popular Natural Language Inference (NLI) benchmark widely used to evaluate cross-lingual Natural Language Understanding (NLU) capabilities across languages. In this paper, we expand XNLI to include Basque, a low-resource language that can greatly benefit from transfer-learning approaches. The new dataset, dubbed XNLIeu, has been developed by first machine-translating the English XNLI corpus into Basque, followed by a manual post-edition step. We have conducted a series of experiments using mono- and multilingual LLMs to assess a) the effect of professional post-edition on the MT system; b) the best cross-lingual strategy for NLI in Basque; and c) whether the choice of the best cross-lingual strategy is influenced by the fact that the dataset is built by translation. The results show that post-edition is necessary and that the translate-train cross-lingual strategy obtains better results overall, although the gain is lower when tested in a dataset that has been built natively from scratch. Our code and datasets are publicly available under open licenses.",
}

APA:

Heredia, M., Etxaniz, J., Zulaika, M., Saralegi, X., Barnes, J., & Soroa, A. (2024). XNLIeu: a dataset for cross-lingual NLI in Basque. In Proceedings of the 2024 Conference of the North American Chapter of the Association for Computational Linguistics: Human Language Technologies (Volume 1: Long Papers) (pp. 4177–4188). Association for Computational Linguistics.