Text Data Augmentation for Deep Learning

Abstract Natural Language Processing (NLP) is one of the most captivating applications Deep Learning. In this survey, we consider how Data Augmentation training strategy can aid in its development. We begin with major motifs summarized into strengthening local decision boundaries, brute force training, causality and counterfactual examples, distinction between meaning form. follow these a concrete list augmentation frameworks that have been developed for text data. Learning generally struggles measurement generalization characterization overfitting. highlight studies cover augmentations construct test sets generalization. NLP at an early stage applying compared to Computer Vision. key differences promising ideas yet be tested NLP. For sake practical implementation, describe tools facilitate such as use consistency regularization, controllers, offline online pipelines, preview few. Finally, discuss interesting topics around task-specific augmentations, prior knowledge self-supervised learning versus Augmentation, intersections transfer multi-task learning, AI-GAs (AI-Generating Algorithms). hope paper inspires further research interest Text Augmentation.