languages:
- en
multilinguality:
- multilingual
paperswithcode_id: xtreme
pretty_name: XTREME
Dataset Card for "xtreme"
Table of Contents
- Dataset Description
- Dataset Structure
- Dataset Creation
- Considerations for Using the Data
- Additional Information
Dataset Description
- Homepage: https://github.com/google-research/xtreme
- Repository: More Information Needed
- Paper: More Information Needed
- Point of Contact: More Information Needed
- Size of downloaded dataset files: 15143.21 MB
- Size of the generated dataset: 1027.42 MB
- Total amount of disk used: 16170.64 MB
Dataset Summary
The Cross-lingual Natural Language Inference (XNLI) corpus is a crowd-sourced collection of 5,000 test and 2,500 dev pairs for the MultiNLI corpus. The pairs are annotated with textual entailment and translated into 14 languages: French, Spanish, German, Greek, Bulgarian, Russian, Turkish, Arabic, Vietnamese, Thai, Chinese, Hindi, Swahili and Urdu. This results in 112.5k annotated pairs. Each premise can be associated with the corresponding hypothesis in the 15 languages, summing up to more than 1.5M combinations. The corpus is made to evaluate how to perform inference in any language (including low-resources ones like Swahili or Urdu) when only English NLI data is available at training time. One solution is cross-lingual sentence encoding, for which XNLI is an evaluation benchmark. The Cross-lingual TRansfer Evaluation of Multilingual Encoders (XTREME) benchmark is a benchmark for the evaluation of the cross-lingual generalization ability of pre-trained multilingual models. It covers 40 typologically diverse languages (spanning 12 language families) and includes nine tasks that collectively require reasoning about different levels of syntax and semantics. The languages in XTREME are selected to maximize language diversity, coverage in existing tasks, and availability of training data. Among these are many under-studied languages, such as the Dravidian languages Tamil (spoken in southern India, Sri Lanka, and Singapore), Telugu and Malayalam (spoken mainly in southern India), and the Niger-Congo languages Swahili and Yoruba, spoken in Africa.
Supported Tasks and Leaderboards
Languages
Dataset Structure
We show detailed information for up to 5 configurations of the dataset.
Data Instances
MLQA.ar.ar
- Size of downloaded dataset files: 72.21 MB
- Size of the generated dataset: 8.77 MB
- Total amount of disk used: 80.98 MB
An example of 'validation' looks as follows.
MLQA.ar.de
- Size of downloaded dataset files: 72.21 MB
- Size of the generated dataset: 2.43 MB
- Total amount of disk used: 74.64 MB
An example of 'validation' looks as follows.
MLQA.ar.en
- Size of downloaded dataset files: 72.21 MB
- Size of the generated dataset: 8.62 MB
- Total amount of disk used: 80.83 MB
An example of 'validation' looks as follows.
MLQA.ar.es
- Size of downloaded dataset files: 72.21 MB
- Size of the generated dataset: 3.12 MB
- Total amount of disk used: 75.33 MB
An example of 'validation' looks as follows.
MLQA.ar.hi
- Size of downloaded dataset files: 72.21 MB
- Size of the generated dataset: 3.17 MB
- Total amount of disk used: 75.38 MB
An example of 'validation' looks as follows.
Data Fields
The data fields are the same among all splits.
MLQA.ar.ar
id: astringfeature.title: astringfeature.context: astringfeature.question: astringfeature.answers: a dictionary feature containing:answer_start: aint32feature.text: astringfeature.
MLQA.ar.de
id: astringfeature.title: astringfeature.context: astringfeature.question: astringfeature.answers: a dictionary feature containing:answer_start: aint32feature.text: astringfeature.
MLQA.ar.en
id: astringfeature.title: astringfeature.context: astringfeature.question: astringfeature.answers: a dictionary feature containing:answer_start: aint32feature.text: astringfeature.
MLQA.ar.es
id: astringfeature.title: astringfeature.context: astringfeature.question: astringfeature.answers: a dictionary feature containing:answer_start: aint32feature.text: astringfeature.
MLQA.ar.hi
id: astringfeature.title: astringfeature.context: astringfeature.question: astringfeature.answers: a dictionary feature containing:answer_start: aint32feature.text: astringfeature.
Data Splits
| name | validation | test |
|---|---|---|
| MLQA.ar.ar | 517 | 5335 |
| MLQA.ar.de | 207 | 1649 |
| MLQA.ar.en | 517 | 5335 |
| MLQA.ar.es | 161 | 1978 |
| MLQA.ar.hi | 186 | 1831 |
Dataset Creation
Curation Rationale
Source Data
Initial Data Collection and Normalization
Who are the source language producers?
Annotations
Annotation process
Who are the annotators?
Personal and Sensitive Information
Considerations for Using the Data
Social Impact of Dataset
Discussion of Biases
Other Known Limitations
Additional Information
Dataset Curators
Licensing Information
Citation Information
@InProceedings{conneau2018xnli,
author = {Conneau, Alexis
and Rinott, Ruty
and Lample, Guillaume
and Williams, Adina
and Bowman, Samuel R.
and Schwenk, Holger
and Stoyanov, Veselin},
title = {XNLI: Evaluating Cross-lingual Sentence Representations},
booktitle = {Proceedings of the 2018 Conference on Empirical Methods
in Natural Language Processing},
year = {2018},
publisher = {Association for Computational Linguistics},
location = {Brussels, Belgium},
}
@article{hu2020xtreme,
author = {Junjie Hu and Sebastian Ruder and Aditya Siddhant and Graham Neubig and Orhan Firat and Melvin Johnson},
title = {XTREME: A Massively Multilingual Multi-task Benchmark for Evaluating Cross-lingual Generalization},
journal = {CoRR},
volume = {abs/2003.11080},
year = {2020},
archivePrefix = {arXiv},
eprint = {2003.11080}
}
Contributions
Thanks to @thomwolf, @jplu, @lewtun, @lvwerra, @lhoestq, @patrickvonplaten, @mariamabarham for adding this dataset.