An Empirical Survey of Data Augmentation for Limited Data Learning in NLP

Abstract NLP has achieved great progress in the past decade through use of neural models and large labeled datasets. The dependence on abundant data prevents from being applied to low-resource settings or novel tasks where significant time, money, expertise is required label massive amounts textual data. Recently, augmentation methods have been explored as a means improving efficiency NLP. To date, there no systematic empirical overview for limited setting, making it difficult understand which work settings. In this paper, we provide an survey recent summarizing landscape (including token-level augmentations, sentence-level adversarial hidden-space augmentations) carrying out experiments 11 datasets covering topics/news classification, inference tasks, paraphrasing single-sentence tasks. Based results, draw several conclusions help practitioners choose appropriate augmentations different discuss current challenges future directions learning