Dialog Management

Abstract

One of the core aspects in the development of conversational interfaces is to design the dialog management strategy. The dialog management strategy defines the system's conversational behaviors in response to user utterances and environmental states. The design of this strategy is usually carried out in industry by handcrafting dialog strategies that are tightly coupled to the application domain in order to optimize the behavior of the conversational interface in that context. More recently, the research community has proposed ways of automating the design of dialog strategies by using statistical models trained with real conversa-tions. This chapter describes the main challenges and tasks in dialog management. We also analyze the main approaches that have been proposed for developing dialog managers and the most important methodologies and standards that can be used for the practical implementation of this important component of a conversa-tional interface. 10.1 Introduction This chapter describes the main aspects, tasks, and approaches involved in the dialog management (DM) process. Section 10.2 defines the DM process and the tasks involved. To illustrate the complexity of dialog strategy design, this section analyzes two frequently arising design issues: the interaction strategy and the choice of a confirmation strategy. Dialog management can be classified into handcrafted approaches using rules, which are described in Sect. 10.3, and statistical approaches using machine learning methodologies, which are described in Sect. 10.4. Statistical approaches have been proposed to model the variability in user behaviors and to allow the exploration of a wider range of strategies. This section provides two detailed examples of the practical application of reinforcement learning and corpus-based supervised learning for the development of statistical dialog managers.