Due to separate levels of representation for descriptions and objects referred to ambiguous or empty reference can be modelled explicitly. Referring expressions typically trigger database requests returning variable numbers of possible referents. The referents are represented in underspecified feature structures separating similarities and differences in the representations. Figure 8 shows an ambiguously referring, a uniquely referring and an overspecified description as they are represented in the third and fourth level of the discourse history as shown in figure 6.
Figure 8: Different representation of descriptions and their referents. The first description refers
ambiguously to a set of objects that is compactly represented in an underspecified feature structure.
The second description refers uniquely to one object. The third description is overspecified; no object
in the data base fits the description. The representations are depicted in a simplified fashion.
In addition to the parallel representation of descriptions and referents, links explicitly represent the relationship between the representations. Since the links are also accessible at the clause level, it is possible to determine for each expression the number of referring objects. This helps us to identify ``what the speaker intends to be picked out by a noun phrase'' [Cohen and Levesque1994]. Consequently, it is possible to formulate dialogue strategies depending on the kind of references found in the discourse.
Moreover, it should be noted that since not only definite descriptions and objects but also actions and goals are represented in feature structures, similar conditions can be formulated on representations of intentions and actions.