Generating Text using Generative Adversarial Networks and Quick-Thought Vectors
David Russell, Longzhuang Li, Feng Tian · 2019
Generative Adversarial Networks (GANs) have been shown to perform very well with the image generation tasks. Many advancements have been made with GANs over the past few years, which are making them more and more accurate in their generation tasks. State-of-the-art methods of natural language processing (NLP) involve word embeddings such as global vectors for word representation (GLoVe) and word2vec. These word embeddings help apply text data to a neural network by converting the textual data to numbers that the networks could use. The main focus for GANs has been image generation and in the past few years there have been research works to apply GANs to the text generation task. This paper presents a Quick-Thought GAN (QTGAN) to generate sentences by incorporating the Quick-Thought model. Quick-Thought vectors offer richer representations than prior unsupervised and supervised methods and enable a classifier to distinguish context sentences from other contrastive sentences. The proposed QTGAN is trained on a portion of the BookCorpus dataset that has been converted to Quick-Thought embeddings. The embeddings generated from the generator are then classified and used to pick a generated sentence. BLEU scores are used to test the results of the training and compared to the Skip-Thought GAN. The increases in BLEU-3 and BLEU-4 scores were achieved with the QTGAN.