Transferable Dialogue Systems and User Simulators

• DOI: 10.48550/arxiv.2107.11904
• Publication Date: 2021-01-01
• Authors: Bo-Hsiang Tseng; Yinpei Dai; Florian Kreyssig; Bill Byrne
• Keywords: Converse; Bootstrapping (finance); Transfer of learning; Domain Adaptation; Labeled data

One of the difficulties in training dialogue systems is lack data. We explore possibility creating data through interaction between a system and user simulator. Our goal to develop modelling framework that can incorporate new scenarios self-play two agents. In this framework, we first pre-train agents on collection source domain dialogues, which equips converse with each other via natural language. With further fine-tuning small amount target data, continue interact aim improving their behaviors using reinforcement learning structured reward functions. experiments MultiWOZ dataset, practical transfer problems are investigated: 1) adaptation 2) single-to-multiple transfer. demonstrate proposed highly effective bootstrapping performance learning. also show our method leads improvements complete datasets.