End-to-End Dialogue System with Multi Languages for Hospital Receptionist Robot

Task-oriented dialogue systems aim to assist users in accomplishing specific tasks with natural language. Recently, with the success of end-to-end chit-chat system, there have been many attempts to build RNN based task-oriented dialogue systems. Hybrid Code Network (HCN) is a practical and efficient end-to-end model using domain-specific software and Recurrent Neural Network (RNN). This paper presents an end-to-end dialogue system for hospital receptionist robot with multi languages using HCN. For this, we synthetically generated dialogue corpus including several tasks. Original HCN was only applied Long Short-Term Memory (LSTM) to train dialogues. We replenish HCN by applying other RNN structures in this architecture such as stacked, reversed input sequence and bidirectional structures. Our proposed RNN model achieved higher performance on the hospital receptionist domain in both English and Korean languages.