In the competition info, there is a rule that said : 'You may use only the datasets provided for this competition.' So in case of data oversampling, it is allowed because you are using the same dataset. But augmenting data in NLP, is quite different than CV, and you need an external data source considering that your train set is very small.
So do you think that using some external sources to augment your data is against the rules here ?
You can oversample your data till the point you're not involving any external data points. Someone correct me if I'm wrong.
By defenition, Oversampling does not involve external data, but in order to get good results with Augmentation, you need an external data source, and here lies the problem.