Response Generation

Abstract

Once the dialog manager has interpreted the user's input and decided how to respond, the next step for the conversational interface is to determine the content of the response and how best to express it. This stage is known as response generation (RG). The system's verbal output is generated as a stretch of text and passed to the text-to-speech component to be rendered as speech. In this chapter, we provide an overview of the technology of RG and discuss tools and other resources. 12.1 Introduction Once the dialog manager has interpreted the user's input and decided how to respond, the next step is to determine the content of the response (content deter-mination) and how best to express it (content realization). Compared with the other components of a conversational interface, there are no readily available tools for response generation (RG). There are two main approaches to RG. In many spoken dialog systems and especially in commercial voice user interfaces, a simple approach is adopted in which the system outputs predetermined responses, using either canned text or templates in which the values of variables can be inserted at runtime. This approach works well in fairly restricted interactions. However, a more elaborate approach is required where the content to be output cannot be determined in advance and where it needs to be edited and cast in a form suitable for spoken output. In this case, techniques from natural language generation (NLG) are required.