Natural language processing course 2022/23: Sentence paraphrasing

Team members:

• Nik Pirnat, [email protected]
• Martin Bavčar, [email protected]
• Anže Glušič, [email protected]

Group public acronym/name: TM9

Enviroment setup

conda create -n nlp-project python=3.8 -c conda-forge
conda activate nlp-project
pip install -r requirements.txt

Preprocessing

Preprocessing the ccKres dataset was done with preprocessing.py.

Back translation

Back-translated dataset was computed using Slovene NMT model with back_translation.py.

Training

Training was run using run_train.py and run_test.py.

Inference

Our models can be downloaded here. Refer to inference.ipynb to run inference on t5-sl-large and t5-sl-small models. Refer to baseline.ipynb for baseline.