Datasets:

HiTZ
/

XNLIvar

license: cc-by-nc-4.0
language:
  - es
  - eu
tags:
  - Variation
pretty_name: xnli_var
configs:
  - config_name: eu
    data_files:
      - split: original
        path:
          - eu/xnli-eu-original.tsv
      - split: native
        path:
          - eu/xnli-eu-native.tsv
      - split: variation
        path:
          - eu/xnli-eu-var.tsv
  - config_name: es
    data_files:
      - split: original
        path:
          - es/xnli-es-original.tsv
      - split: native
        path:
          - es/xnli-eu2es-native.tsv
      - split: variation
        path:
          - es/xnli-es-var.tsv
  - config_name: other_ablation
    data_files:
      - split: Basque_no_repetition
        path:
          - ablation-eu/xnli-native-var-eu-NO-REPETITION.tsv
      - split: Basque_less_western
        path:
          - ablation-eu/xnli-native-var-eu-less-biz.tsv
      - split: Basque_less_central
        path:
          - ablation-eu/xnli-native-var-eu-less-gip.tsv
      - split: Spanish_no_repetition
        path:
          - ablation-es/xnli-native-var-es-no-rep.tsv
  - config_name: translations
    data_files:
      - split: translations_Basque_native
        path:
          - translations/xnli-eu2en-native.tsv
      - split: translations_Basque_variation
        path:
          - translations/xnli-eu2en-var.tsv
      - split: translation_Spanish_native
        path:
          - translations/xnli-es2en-native.tsv
      - split: translation_Spanish_variation
        path:
          - translations/xnli-es2en-var.tsv

XNLIvar: Basque and Spanish variation-inclusive NLI

This repository contains the data and code used in the paper Lost in Variation? Evaluating NLI Performance in Basque and Spanish Geographical Variants

This paper evaluates the capacity of current language technologies to understand Basque and Spanish language varieties. We use Natural Language Inferenc (NLI) as a pivot task and introduce a novel, manually-curated parallel dataset in Basque and Spanish and their corresponding variants. Empirical analysis of comprehensive crosslingual and in-context learning experiments with, respectively, encoder-only and decoder-based Large Language Models (LLMs), reveals a performance drop when processing linguistic variations, with more pronounced effects observed in Basque. Error analysis indicates that lexical overlap plays no role, suggesting that linguistic variation represents the primary reason for the lower results. All data and code in this repository are public under Attribution-NonCommercial 4.0 International license.

Dataset decription

XNLIeu (the origin dataset for XNLIvar) has been derived from XNLI and it is distributed under its same license.

XNLIvar, a novel, manually-curated, variation inclusive NLI datasets in Basque and Spanish for NLI evaluation. The data is structured in the following folders:

eu: It provides the original XNLI test data, as well as the native and variation inclusive datasets.
es: It provides de original XNLI test data, as well as the Basque native data translated into Spanish and the variation inclusive data.
Translations: It provides the automatic translations of the native and variation inclusive datasets into English.
ablation-eu: The datasets used in the ablation experiments for Basque.
ablation-es: The datasets used in the ablation experiments for Spanish.

Useful links

Citation

The paper that explains the dataset and experiments can be cited as follows:

@inproceedings{bengoetxea-et-al-2025,,
    title = "Lost in Variation? Evaluating NLI Performance in Basque and Spanish Geographical Variants",
    author = "Bengoetxea, Jaione  and
      Gonzalez-Dios, Itziar  and
      Agerri, Rodrigo",
      year = "2025",
      url = "https://arxiv.org/abs/2506.15239"
}