Data Augmentation for Text Generation Without Any Augmented Data
This addresses the challenge of designing effective data augmentation functions for text generation models, offering a more automated approach.
The paper tackles the problem of data augmentation for text generation by deriving an objective that eliminates the need for manually defined mapping functions to create augmented data, and experiments on five datasets show it can approximate or surpass popular augmentation methods.
Data augmentation is an effective way to improve the performance of many neural text generation models. However, current data augmentation methods need to define or choose proper data mapping functions that map the original samples into the augmented samples. In this work, we derive an objective to formulate the problem of data augmentation on text generation tasks without any use of augmented data constructed by specific mapping functions. Our proposed objective can be efficiently optimized and applied to popular loss functions on text generation tasks with a convergence rate guarantee. Experiments on five datasets of two text generation tasks show that our approach can approximate or even surpass popular data augmentation methods.