We chose as the basic representation formalism throughout the system typed feature structures [Carpenter1992]. The types are ordered in a conceptual model, the type hierarchy, which represents domain-specific as well as domain-independent terminological knowledge using IS-A and IS-PART-OF relations. Figure 1 shows a schematic view of part of the type hierarchy we use in our interactive map application.
There are several small domain-specific sub-models for semantically closed domains. Among these are hierarchies introducing concepts of time, days and dates, or reservations, or objects that can be displayed on a map. In addition, there are domain models representing different speech acts, gestures in case of multimodal input and so on. These domain models are domain-independent. The domain model for one particular application is then combined with several domain-dependent sub-models and the domain-independent model. In addition, there is one particular type hierarchy declaring the information necessary for the application to perform the goals. The junction of all type hierarchies is subsequently referred to as the domain model . The domain model answer the first of the three questions, namely which are the entities, properties and actions in the domain and how do they relate to each other.
Note that since the domain model is a type hierarchy, and as such allows techniques such as inheritance, reasoning (such as reasoning based on the questions if the goal has been determined uniquely) about the nature of the goal may take place without knowing what specifically the goal is. This fact is the computational basis that allows us to express dialogue strategies in a domain-independent way, while retaining the possibility of overloading goal execution operators with domain-specific procedures.