In this section, we will illustrate the interaction of knowledge sources as specified by the clauses. We first look at some specific processing steps before providing an example of a full dialogue.
A user's request, e.g. I would like to reserve a table may be mapped, due to recognition errors and partial parsing, to the following partial representation
The two matching task descriptions would be the one for hotel reservation and restaurant reservation,
the corresponding underspecified feature structure representing both descriptions would have the value
of the path 
set to 
which would
prompt a corresponding clarification question. Subsequent unification with the semantic representation
of the answer a restaurant reservation please will disambiguate entirely the representation
on top of the users' stack. Since now the intention of the user has been determined,
clauses calculating the informational differences between the information required in the task description
and the information available in the discourse fire to obtain complementary information.
In this case, the system will prompt for the arrival date. The communicative goal of this action is to obtain
the specified information, consequently, a representation of the goal is pushed onto the stack G and
a semantic representation containing the propositional content of the question is generated. This
leads to the situation depicted in figure 7. The user's answer the day after tomorrow
generates a semantic representation of the form 
which, in turn,
will trigger a database lookup, unifying a representation of the actual date with the representation
of the deictic expression. Since this information is more specific than the communicative goal, the
sub dialogue is closed and the new information is integrated in the representation of the intention of the
system. The requirement for the path 
is thus fulfilled, and another path is
selected in order to obtain complementary information.
Note that the user's response can also generate a sub dialogue. If for example the systems question Which category would you prefer is answered with How much is the cheapest, the incoming information is not compatible with the communicative goal and, moreover, is not a repair, so a new subdialogue is entered. in this way, hierarchical dialogue structure is inferred.
An utterance like i don't need a reservation any more will generate a representation of the form
This representation will cause the comparison of the value of the 
feature in the semantic representation with the value of the 
feature of the currently activated
task description on the stack to fail. This, in turn, will cause
this task description and all descriptions on top of it to be removed.
After having examined some details of the dialogue processing, we now present an example of a full dialogue in figure 10.
: An example for a dialogue.