To develop a goal-oriented slot-filling dialogue system, we therefore attempt to understand the user utterance based on two subproblems: DA and NE extractions.įigure 8.2. Figure 8.2(a), for example, shows the semantic frame structure, based on an XML representation, for an air travel information service.

In this chapter we define NE as the domain-specific semantic meaning of a word. The NE is an identifier of an entity such as a person, location, organization, or time. The DA represents the meaning of an utterance at the discourse level, which is approximately the equivalent of the intent or subject slot in practical dialogue systems. This frame is usually divided into two main components: dialogue act (DA) and named entity (NE).

To understand the user utterance, most SLU systems define a semantic frame-that is, a formal structure of predicted meanings consisting of slot/value pairs. The goal of SLU for slot-filling dialogue applications is to extract meanings from natural language speech and to infer the speaker's intention.

Tcpvcon