Dialogue Generation Using Self-Attention Generative Adversarial Network

Developing an intelligent conversation system is one of the longest-running goals in Artificial Intelligence research. For a successful conversation system, appropriate dialogue generation is one of the main components. This article describes a method to generate dialogue using Self-Attention Generative Adversarial Network (SAGAN). The general objective of this method is to produce dialogue which is very similar to human-generated dialogue both structurally and contextually. Generative Adversarial Network (GAN) is mainly used to generate dialogue which is structurally very similar to human-generated dialogue. On the other hand, the self-attention network helps to maintain the context of the conversation in extreme detail for both single track and multi-track dialogue generation. SAGAN enhances the accuracy for both single track and multi-track dialogue generation in a significant way.