overlim / README.md
Robin Kurtz
readme
99ccef4
|
raw
history blame
3.18 kB
metadata
annotations_creators:
  - other
language_creators:
  - other
languages:
  - sv
  - da
  - nb
licenses:
  - cc-by-4.0
multilinguality:
  - translation
pretty_name: overlim
size_categories:
  - unknown
source_datasets:
  - extended|glue
  - extended|super_glue
task_categories:
  - text-classification
  - text-scoring
task_ids:
  - natural-language-inference
  - semantic-similarity-classification
  - sentiment-classification
  - text-classification-other-paraphrase-identification
  - text-classification-other-qa-nli

Dataset Card for OverLim

Dataset Description

  • Homepage:
  • Repository:
  • Paper:
  • Leaderboard:
  • Point of Contact:

Dataset Summary

The OverLim dataset contains some of the GLUE and SuperGLUE tasks automatically translated to Swedish, Danish, and Norwegian (bokmål), using the OpusMT models for MarianMT.

The translation quality was not manually checked and may thus be faulty. Results on these datasets should thus be interpreted carefully.

Supported Tasks and Leaderboards

The data contains the following tasks from GLUE and SuperGLUE:

  • GLUE
    • mnli
    • mrpc
    • qnli
    • qqp
    • rte
    • sst
    • stsb
    • wnli
  • SuperGLUE
    • boolq
    • cb
    • copa
    • rte

Languages

  • Swedish
  • Danish
  • Norwegian (bokmål)

Dataset Structure

Data Instances

Every task has their own set of features, but all share an idx and label.

  • GLUE
    • mnli
      • premise, hypothesis
    • mrpc
      • text_a, text_b
    • qnli
      • premise, hypothesis
    • qqp
      • text_a, text_b
    • sst
      • text
    • stsb
      • text_a, text_b
    • wnli
      • premise, hypothesis
  • SuperGLUE
    • boolq
      • question, passage
    • cb
      • premise, hypothesis
    • copa
      • premise, choice1, choice2, question
    • rte
      • premise, hypothesis

Data Splits

In order to have test-split, we repurpose the original validation-split as test-split, and split the training-split into a new training- and validation-split, with an 80-20 distribution.

Dataset Creation

For more information about the individual tasks see (https://gluebenchmark.com) and (https://super.gluebenchmark.com).

Curation Rationale

Training non-English models is easy, but there is a lack of evaluation datasets to compare their actual performance.

Source Data

Initial Data Collection and Normalization

[More Information Needed]

Who are the source language producers?

[More Information Needed]

Annotations

Annotation process

[More Information Needed]

Who are the annotators?

[More Information Needed]

Personal and Sensitive Information

[More Information Needed]

Considerations for Using the Data

Social Impact of Dataset

[More Information Needed]

Discussion of Biases

[More Information Needed]

Other Known Limitations

[More Information Needed]

Additional Information

Dataset Curators

[More Information Needed]

Licensing Information

[More Information Needed]

Citation Information

[More Information Needed]

Contributions

Thanks to @robinqrtz for adding this dataset.