The requisites of a powered-AI system is to have a big enough annotated data. Lack of the datasets is a big challenge to obtain the robustness of the predictive models so that it can broaden the AI ideas to various domains. The predictive models are less generalized and prone to overfit. Although the resources for Vietnamese have been investigated more and more, it has still been a low-resources language which is the biggest barrier in order to leverage the robustness of the AI applications. Building the datasets consumes so much time and money. This paper presents the text augmentation to generate the new annotated training data without user's intervention. This paper has summarized the potential methods, especially for the cross-languages methods to enhance the data, analyzed and evaluated the advantages and disadvantages of each method to apply to Vietnamese language processing. The synthetic presentation shows text augmentation has gained competitive performances and helped to save the time and money to build the data.