We use typed feature structures [Carpenter1992] such as the one shown in figure 2 to represent the semantics of the users' requests. Each structure represents the semantics of a phrase of one of the main syntactic categories NP, VP, or PP. Feature structures are particularly well-suited for dialogue processing since partial information may be modelled adequately. This allows for easy integration of additional knowledge bases. As an example, consider the result of a database request filling out a partially instantiated feature structure.
Figure 2: An example of a typed feature structure representing a request to show a museum.
Since the feature structures are typed we can use them to express anything from definite descriptions, to speech acts and intentions and goals. This allows us to perform any actions, such as unification, compatibility check or disambiguation, on representations of speech acts and intentions in the same way as we do on representations of objects.