The Cross-lingual Choice of Plausible Alternatives dataset is a benchmark to evaluate the ability of machine learning models to transfer commonsense reasoning across languages. The dataset is the translation and reannotation of the English COPA (Roemmele et al. 2011) and covers 11 languages from 11 families and several areas around the globe. The dataset is challenging as it requires both the command of world knowledge and the ability to generalise to new languages. All the details about the creation of XCOPA and the implementation of the baselines are available in the paper.


The XCOPA data are stored in data. Alternatively, they can be loaded through 🤗Datasets as follows:

from datasets import load_dataset

xcopa_dataset = load_dataset('xcopa')

The "translate test" data obtained via Google Translate are available in the folder data-gmt (note that these do not include Quechua).


Language Premise Question Choice 1 Choice 2
qu Sipasqa cereal mikhunanpi kuruta tarirqan. Result Payqa pukunman ñuqñuta churakurqan. Payqa manam mikhuyta munarqanchu.
en The girl found a bug in her cereal. Result She poured milk in the bowl. She lost her appetite.
th ตาของฉันแดงและบวม Cause ฉันร้องไห้ ฉันหัวเราะ
en My eyes became red and puffy. Cause I was sobbing. I was laughing.


ISO 639-2 Name Family Area1
et Estonian Uralic Northern Europe
ht Haitian Creole French Creole Carribean
id Indonesian Austronesian Southeastern Asia
it Italian Indo-European Southern Europe
qu Southern Quechua2 Quechuan Southern America
sw Swahili Niger-Congo Eastern Africa
ta Tamil Dravidian Southern Asia
th Thai Kra-Dai Southeastern Asia
tr Turkish Turkic Western Asia
vi Vietnamese Austroasiatic Southeastern Asia
zh Mandarin Chinese Sino-Tibetan Eastern Asia

1 According to the United Nations geoscheme.

2 Translation by Irma Alvarez Ccoscco, an Eastern Apurímac Quechua speaker.


If you want to see your scores reported, please make a pull request changing the table below.

Model Paper et ht id it qu sw ta th tr vi zh
Human Ponti et al. (2020) 98.2 96.4 100.0 97.0 94.8 99.0 98.6 98.2 96.4 98.4 96.6
RoBERTa Large (Translate test) Ponti et al. (2020) 81.0 73.8 82.2 77.8 - 74.2 79.6 71.4 79.6 81.0 86.0
XLM-R Large Ponti et al. (2020) 71.4 - 79.8 72.6 - 59.2 73.0 72.8 74.4 73.8 78.6
MAD-X Base Pfeiffer et al. (2020) 61.3 53.7 65.8 63.0 52.5 56.3 61.9 61.8 60.3 66.1 67.6

The performance of other multilingual pre-trained encoders is shown in the figure.


If you use the data from this repository, please cite both XCOPA \cite{ponti2020xcopa} and the original COPA paper \cite{roemmele2011choice}.

  title={{XCOPA: A} Multilingual Dataset for Causal Commonsense Reasoning},
  author={Edoardo M. Ponti, Goran Glava\v{s}, Olga Majewska, Qianchu Liu, Ivan Vuli\'{c} and Anna Korhonen},
  booktitle={Proceedings of the 2020 Conference on Empirical Methods in Natural Language Processing (EMNLP)},

  title={Choice of plausible alternatives: An evaluation of commonsense causal reasoning},
  author={Roemmele, Melissa and Bejan, Cosmin Adrian and Gordon, Andrew S},
  booktitle={2011 AAAI Spring Symposium Series},