A Challenge on Semi-Supervised and Reinforced Task-Oriented Dialog Systems

Task-oriented dialogue (TOD) systems are designed to assist users to accomplish their goals, and have gained more and more attention in both academia and industry with recent advances in neural approaches (Williams et al., 2016; Gao et al., 2019). A TOD system typically consists of several modules, which track user goals to update dialog states, query a task-related knowledge base (KB) using the dialog states, decide actions and generate responses. Unfortunately, building TOD systems remains a label-intensive, time-consuming task for two main reasons.