Skip to content
Switch branches/tags

Latest commit


Git stats


Failed to load latest commit information.
Latest commit message
Commit time

UnsupNTS: Unsupervised Neural Text Simplification

This is the original implementation of the Unsupervised Neural Text Simplification system and their semi-supervised variants mentioned in the ACL 2019 long paper:

Sai Surya, Abhijit Mishra, Anirban Laha, Parag Jain, and Karthik Sankaranarayanan. Unsupervised Neural Text Simplification arXiv preprint arXiv:1810.07931 (2018).


Download from link and extract

unzip has

  • Unsupervised sets of easy and difficult set of sentences judged on readability ease scores.
  • Dict2vec embeddings trained on the above unsupervised sets.
  • 10k parallel pairs of difficult and simplified variants.
  • Test set and references - eight tab seperated references per each test sentence.

Train the models using

bash has

  • UNTS system from unsupervised simplification data using the exact same settings described in the paper.
  • UNTS-10k system, using additional 10k supervised pairs of mixture of split-rephrase and simplification parallel pairs.
  • UNMT system on the unsupervised simplification data.
  • ablations on adversarial and separation/classifier losses.

For more details and additional options, run the above scripts with the --help flag. Alternatively, visit the ipynb in google colaboratory to reproduce the results. To access pretrained models visit link. The folder predictions has the generations from the pretrained models.

Note: Pretrained models were trained with pytorch 0.3.1.

Generation and Evaluation of Simplifications

bash is used for

  • Generating simplifications of test dataset.
  • Computing stand alone metrics such as Flesch readability ease score difference, Tree similarity and Document similarity metrics.
  • Computing SARI, BLEU and Word-diff metrics.


Our code uses functions from and extensively.

If you use our system for academic research, please cite the following paper:

    title = "Unsupervised Neural Text Simplification",
    author = "Surya, Sai  and
      Mishra, Abhijit  and
      Laha, Anirban  and
      Jain, Parag  and
      Sankaranarayanan, Karthik",
    booktitle = "Proceedings of the 57th Annual Meeting of the Association for Computational Linguistics",
    month = jul,
    year = "2019",
    address = "Florence, Italy",
    publisher = "Association for Computational Linguistics",
    url = "",
    doi = "10.18653/v1/P19-1198",
    pages = "2058--2068"


Unsupervised Neural Text Simplification







No releases published


No packages published