ekgren/swedish_nlp_datasets

Swedish NLP dataset resources

Swedish NLP dataset resources

Here we collect resources for Swedish Text Data. We are looking for collaborators to add sources and scraped data.

Raw text datasets

• Swedish wikipedia Wikipedia is a great resource for high quality text.
  First install tools with ´install-tools.sh´ then run ´./get-data-wiki.sh sv´ and it will download and preprocess the swedish wikipedia data

• Litteraturbanken

• Oscar

Evaluation datasets

• SUPERLIM

• XNLI for Swedish
  https://github.com/salesforce/xnli_extension
  Translated from English to Swedish with Google Translate
  Paper: BERT is Not an Interlingua and the Bias of Tokenization [1]

• STS-b for Swedish
  https://github.com/timpal0l/sts-benchmark-swedish
  Translated from English to Swedish with Google Translate
  Paper: Why Not Simply Translate? A First Swedish Evaluation Benchmark for Semantic Similarity [2]

• Swedish reviews
  https://github.com/huggingface/datasets/tree/master/datasets/swedish_reviews

Swedish datasets on huggingface datasets

• There are a lot of Swedish datasets on huggingface datasets

General Swedish corpora resources

• https://www.ling.su.se/english/nlp/corpora-and-resources
• https://spraakbanken.gu.se/en/resources

References

[1]

@article{singhMultiLingTokBias2019,
title={{BERT is Not an Interlingua and the Bias of Tokenization}},
author={Singh, Jasdeep and McCann, Bryan and Xiong, Caiming and Socher, Richard},
journal={The Workshop on Deep Learning for Low-Resource NLP at EMNLP 2019},
year={2019}
}

[2]

@article{isbister2020not,
  title={Why Not Simply Translate? A First Swedish Evaluation Benchmark for Semantic Similarity},
  author={Isbister, Tim and Sahlgren, Magnus},
  journal={arXiv preprint arXiv:2009.03116},
  year={2020}
}