metadata

pretty_name: ExtraGLUE
language:
  - pt
source_datasets:
  - glue
  - superglue
license: mit
viewer: false
task_categories:
  - text-classification
  - sentence-similarity
  - question-answering
task_ids:
  - language-modeling
  - multi-class-classification
  - natural-language-inference
  - sentiment-classification
  - semantic-similarity-scoring
  - semantic-similarity-classification

This is the dataset card for extraGLUE. You may be interested in some of the other datasets for Portuguese and in the models trained with them, namely Albertina (encoders) and Gervásio (decoders) families.

ExtraGLUE

ExtraGLUE is a Portuguese dataset obtained by the automatic translation of some of the tasks in the GLUE and SuperGLUE benchmarks. Two variants of Portuguese are considered, namely European Portuguese and American Portuguese. The dataset is distributed for free under an open license.

The 14 tasks in extraGLUE cover different aspects of language understanding:

Single sentence

SST-2 is a task for predicting the sentiment polarity of movie reviews.

Semantic similarity

MRPC is a task for determining whether a pair of sentences are mutual paraphrases.
STS-B is a task for predicting a similarity score (from 1 to 5) for each sentence pair.

Inference

MNLI is a task to determine if a given premise sentence entails, contradicts, or is neutral to a hypothesis sentence; this task includes matched (in-domain) and mismatched (cross-domain) validation and test sets.
QNLI is a question-answering task converted to determine whether the context sentence contains the answer to the question.
RTE is a task for determining whether a premise sentence entails a hypothesis sentence.
WNLI is a pronoun resolution task formulated as sentence pair entailment classification where, in the second sentence, the pronoun is replaced by a possible referent.
CB comprises short texts with embedded clauses; one such clause is extracted as a hypothesis and should be classified as neutral, entailment or contradiction.
AX_b is designed to test models across a wide spectrum of linguistic, commonsense, and world knowledge; each instance contains a sentence pair labeled with entailment or not entailment.
AX_g is designed to measure gender bias, where each premise sentence includes a male or female pronoun and a hypothesis includes a possible referent for the pronoun.

Question answering

BoolQ is a question-answering task where yes/no questions are given for short text passages.
MultiRC is a task where, given a context paragraph, a question, and an answer, the goal is to determine whether the answer is true; for the same context and question, more than one answer may be correct.

Reasoning

CoPA is a casual reasoning task: given a premise, two choices, and a cause/effect prompt, the system must choose one of the choices.

If you use this dataset please cite:

@misc{osório2024portulan,
  title={PORTULAN ExtraGLUE Datasets and Models: Kick-starting a Benchmark for the Neural Processing of Portuguese},
  author={Tomás Osório and Bernardo Leite and Henrique Lopes Cardoso and Luís Gomes and João Rodrigues and Rodrigo Santos and António Branco},
  year={2024},
  eprint={2404.05333},
  archivePrefix={arXiv},
  primaryClass={cs.CL}
}

Acknowledgments

The research reported here was partially supported by: PORTULAN CLARIN—Research Infrastructure for the Science and Technology of Language, funded by Lisboa 2020, Alentejo 2020 and FCT—Fundação para a Ciência e Tecnologia under the grant PINFRA/22117/2016; research project GPT-PT - Transformer-based Decoder for the Portuguese Language, funded by FCT—Fundação para a Ciência e Tecnologia under the grant CPCA-IAC/AV/478395/2022; innovation project ACCELERAT.AI - Multilingual Intelligent Contact Centers, funded by IAPMEI, I.P. - Agência para a Competitividade e Inovação under the grant C625734525-00462629, of Plano de Recuperação e Resiliência, call RE-C05-i01.01 – Agendas/Alianças Mobilizadoras para a Reindustrialização; and Base Funding (UIDB/00027/2020) and Programmatic Funding (UIDP/00027/2020) of the Artificial Intelligence and Computer Science Laboratory (LIACC) funded by national funds through FCT/MCTES (PIDDAC).