Code release for submission made to GEM-Benchmark 2021 Text Simplification Shared-Task on TurkCorpus ans ASSET datasets
- Python >= 3.7
git clone https://github.com/kvadityasrivatsa/gem_2021_simplification_task.git
cd gem_2021_simplification_task
./install.sh
Train the submission model on WikiLarge
- for TurkCorpus:
python3 train.py --evalset turk --ner --nbchars 0.95 --levsim 0.75 --wrdrank 0.75
- for ASSET:
python3 train.py --evalset asset --ner --nbchars 0.95 --levsim 0.75 --wrdrank 0.75
Generate and evaluate output (on SARI score)
- for TurkCorpus:
python3 evaluate.py --evalset turk
- for ASSET:
python3 evaluate.py --evalset asset
The checkpoint for our model with the best scores is available here
(Note: The official system-desciption for the model can be found here)
Our model builds upon the ACCESS model proposed in Controllable Sentence Simplification (Martin et al., 2020).
- KV Aditya Srivatsa (k.v.aditya@research.iiit.ac.in)
- Monil Gokani (monil.gokani@research.iiit.ac.in)
If you have any queries, please do reach out.
Refer to the LICENSE file for more details.