Neural Dialogue System with Emotion Embeddings

Emotional intelligence is a vital human mechanism that allows people to identify and react to different feelings in various environments, especially conversations. That is why it is important to address the emotional aspect of generative dialogue systems. We propose to condition a neural dialogue system - a deep recurrent neural network - on response emotion, training emotion embeddings in a similar fashion to word embeddings. This way, during the response generation phase, the system can generate not only content-wise appropriate, but also having required emotional tone, utterances.