Natural language processing course 2022/23: Paraphrasing sentences

Team members:

• Nela Petrželková, 70090038, [email protected]
• Žan Horvat, 63190120, [email protected]
• Žiga Patačko Koderman, 63200339, [email protected]

Group public acronym/name: Krote

---

Abstract: Sentence paraphrasing is a widely studied problem in natural language processing. In this paper, we discuss the problem of sentence paraphrasing and compare different approaches to this task. We introduce different evaluation metrics that are used to assess the quality of the models created. We present our approach to sentence paraphrasing and compare it with a model based on traditional methods. We also introduce our data set that was used to the model training. Our data set was generated using OpenAI's GPT-3 model, which is a novel approach to data set generation, and we also compare it to the traditional back-translation method.

The final report can be found in folder 3 - submission.

The fine-tuned models can be downloaded:

• Model trained for 384 epochs on data generated using backtranslation is available here.
• Model trained for 128 epochs on data generated using OpenAI's Davinci model is available here.