Augmentation

When there is not enough labeled text, you can make more of it from the text you have. The page has one overview of the options, from synonym swaps and embedding neighbors to back translation, contextualized word embeddings, and text generation, plus a pointer to a second post on the same topic.

The same notes are in Embedding, LANGUAGE TRANSLATION, and Synonyms.

The overview is the Medium post on data augmentation in NLP. It starts with Synonyms, similar embedded words (w2v), back translation, contextualized word embeddings, text generation, each a step further from the original sentence. Yonatan Hadar also has a medium post about this.

The Towards Data Science copy of the same post, covering synonyms, similar embedded words (w2v), back translation, contextualized word embeddings, and text generation, is https://towardsdatascience.com/data-augmentation-in-nlp-2801a34dfc28.