Data sparsity is one of the key challenges associated with model development\nin Natural Language Understanding (NLU) for conversational agents. The\nchallenge is made more complex by the demand for high quality annotated\nutterances commonly required for supervised learning, usually resulting in\nweeks of manual labor and high cost. In this paper, we present our results on\nboosting NLU model performance through training data augmentation using a\nsequential generative adversarial network (GAN). We explore data generation in\nthe context of two tasks, the bootstrapping of a new language and the handling\nof low resource features. For both tasks we explore three sequential GAN\narchitectures, one with a token-level reward function, another with our own\nimplementation of a token-level Monte Carlo rollout reward, and a third with\nsentence-level reward. We evaluate the performance of these feedback models\nacross several sampling methodologies and compare our results to upsampling the\noriginal data to the same scale. We further improve the GAN model performance\nthrough the transfer learning of the pretrained embeddings. Our experiments\nreveal synthetic data generated using the sequential generative adversarial\nnetwork provides significant performance boosts across multiple metrics and can\nbe a major benefit to the NLU tasks.\n