Data Augmentation in NLP

You might have optimal machine learning algorithm to solve your problem. But once you apply it in real world soon you will realize that you need to train it on more data. Due to lack of large dataset you will try to further optimize the algorithm, tune hyper-parameters or look for some low tech approach. Most state of the art machine learning models are trained on large datasets. Real world performance of machine learning solutions drastically improves with more data.