Text rephrasing

Is there a profound method for rephrasing existing texts (e.g. for data augmentation)? I have already tried to translate the text into another language and retranslate but this changes the text only a little bit.

1 Like

You could use the T5 or the Pegasus models for paraphrasing, hope you also tried parrot paraphraser which uses T5 in the backend. You can also use google translate api for back translation, ideally using languages of different origin (ex: Spanish, Tamil, German) would give different results. Also nlpaug is a good tool for doing multiple augmentations.

2 Likes

As @imrandude pointed out, focus on paraphrasing. In production, there are some cool tools like quillbot which just does that.

https://quillbot.com/

Hi Thanks! The Pegasus Model worked better for me than the T5 Parrot one

1 Like

MarianMT > Pegasus > T5 for translation based on my experience.

In the Hub, these models have the prefix “Helsinki-”

1 Like