Text Data Augmentation for the Korean Language

Data augmentation (DA) is a universal technique to reduce overfitting and improve the robustness of machine learning models by increasing quantity variety training dataset. Although data essential in vision tasks, it rarely applied text datasets since less straightforward. Some studies have concerned augmentation, but most them are for majority languages, such as English or French. There been only few on minority e.g., Korean. This study fills gap demonstrating several common methods Korean corpora with pre-trained language models. In short, we evaluate performance two approaches, known transformation back translation. We compare these augmentations among four downstream tasks: semantic textual similarity (STS), natural inference (NLI), question duplication verification (QDV), sentiment classification (STC). Compared cases without gains when applying 2.24%, 2.19%, 0.66%, 0.08% STS, NLI, QDV, STC respectively.