The project gathered a large dataset of Finnish paraphrase pairs (over 100,000) accompanied by a small test set of Swedish paraphrases. The paraphrases are selected and classified manually, so as to minimize lexical overlap, and provide examples that are maximally structurally and lexically different. The objective is to create a dataset which is challenging and better tests the capabilities of natural language understanding. An important feature of the data is that most paraphrase pairs are distributed in their document context. The primary application for the dataset is the development and evaluation of deep language models, and representation learning in general. The project is funded by the European Language Grid (2020-2021) and the Academy of Finland (2021-2023).

Data and model download

• The full manually annotated dataset is on GitHub and the European Language Grid
• A model trained on this data to classify text segment pairs into the various classes of paraphrase is documented and downloadble in the following notebook [GitHub] [Colab]
• A collection of paraphrase candidates (500K positive and 5M negative) gathered using the above model and useful for further training in a straightforward classification setup is downloadable here
• Finnish SentenceBERT models fine-tuned on the data are available for download here. They can also be accessed using Huggingface with the model names TurkuNLP/sbert-cased-finnish-paraphrase and TurkuNLP/sbert-uncased-finnish-paraphrase, where example code snippets are provided. Check out the DEMO.

Documentation and publications

• An initial release of the data is described in this Nodalida’21 paper
• The annotation guidelines of the data are on arxiv
• The data files are documented in the README file of the data release
• Project poster

Annotation tools and other software

• Annotation tools: candidate selection, candidate annotation, statistics

How to cite?

Main reference:

J. Kanerva & F. Ginter & LH. Chang & I. Rastas & V. Skantsi & J. Kilpeläinen & HM. Kupari & A. Piirto & J. Saarni & M. Sevón & O. Tarkka. 2023. Towards Diverse and Contextually Anchored Paraphrase Modeling: A Dataset and Baselines for Finnish. Natural Language Engineering. 2024;30(2):319-353. doi:10.1017/S1351324923000086.

@article{kanerva2023paraphrasejournal,
    author={Jenna Kanerva and Filip Ginter and Li-Hsin Chang and Iiro Rastas and Valtteri Skantsi and Jemina Kilpel{\"a}inen and Hanna-Mari Kupari and Aurora Piirto and Jenna Saarni and     Maija Sev{\'o}n and Otto Tarkka},
    title={Towards Diverse and Contextually Anchored Paraphrase Modeling: A Dataset and Baselines for {Finnish}},
    year={2023},
    journal={Natural Language Engineering},
    publisher={Cambridge University Press},
    doi = {10.1017/S1351324923000086},
    url = {https://doi.org/10.1017/S1351324923000086}
}

Other publications:

J. Kanerva & F. Ginter & LH. Chang & I. Rastas & V. Skantsi & J. Kilpeläinen & HM. Kupari & J. Saarni & M. Sevón & O. Tarkka. 2021. Finnish Paraphrase Corpus. Proceedings of the 23rd Nordic Conference on Computational Linguistics (NoDaLiDa 2021) pp. 288-298.

J. Kanerva & H. Kitti & LH. Chang & T. Vahtola & M. Creutz & F. Ginter. 2024. Semantic search as extractive paraphrase span detection. Language Resources and Evaluation. https://doi.org/10.1007/s10579-023-09715-7

Acknowledgements

• Project team: Li-Hsin Chang, Filip Ginter, Jenna Kanerva, Jemina Kilpeläinen, Hanna-Mari Kupari, Aurora Piirto, Iiro Rastas, Jenna Saarni, Maija Sevón, Valtteri Skantsi, Otto Tarkka (alphabetic order)
• Funding:
  • European Language Grid
  • Academy of Finland
• Data:
  • opensubtitles.org