Generative Adversarial Approach in Natural Language Processing

Abstract

The use of a generative adversarial algorithm for training neural networks made it possible to make significant progress in solving the problem of generating images and audio data. Nevertheless, important problems remain in solving the tasks of generating discrete data sequences. Solving such problems will allow using generative-adversarial learning to generate text data. This paper reflects a brief overview of modern research and achievements in the generation of text data using generative adversarial learning, lists a set of tasks that can be solved using this approach, describes possible problems and existing methods for solving existing problems, and also describes some suggestions for improving models. The structure and algorithm of the proposed system are described, the research results are presented.