Does Data Augmentation Improve Generalization in NLP?

Rohan Jha, Charles Lovering, Ellie Pavlick

Neural models often exploit superficial features to achieve good performance, rather than deriving more general features. Overcoming this tendency is a central challenge in areas s…