Introduction of the information from the CMs into the UM

CMs represent the current hypotheses on the user's plans and goals. Since not always users are explicit in manifesting their goals, several CMs are usually active at the same time, representing alternative views on their intentions. Carberry (see [Carberry1988]) specifies that the CMs are part of the UM. However, in her framework the actions in the active CMs are are not used for making further inferences on the contents of the UM in other way than by extending the CMs themselves (by means of expansion procedures, like the upward expansion, or by attaching a new recognized action to the active CMs with the focusing procedure). In our opinion, although active CMs represent the intentions attributed by the system to users and contribute to their description, taking all of them as part of the UM would cause a combinatorial explosion of the hypotheses space. On the other hand, even when several CMs are active, the fact that they are built from the same set of utterances makes very likely that some actions occur in each of them. In this case, we think that the intention to execute such actions can be added to the UM, because it represents a single hypothesis, belonging to every interpretation of the ongoing dialogue. This addition makes it possible to extend the UM by means of new inferences, without introducing any ambiguity.

The basic procedure for the selection of the intentions to be included into the UM is the select-common-actions, defined in the following, that is applied after the execution of the upward expansion phase of the dialogue interpretation algorithm.

select-common-actions (CM-SET):

 1) find the most specific actions  in the GH such that for each 
element in

		 CM-SET there exists an action  that is subsumed by 

2) for all  do

		 a) for each parameter p of 

		 		 identify the set of restrictions that subsume all the restrictions 
associated

		 		 to p in each of the occurrences of  in CM-SET

		 b) add to UM the entry 

		 		 where  is obtained from  by associating to each of its 
parameters the

		 		 set of restrictions identified in step 2a)

		 c) if UM  already contains an intention I relative to an instance 
 of A

		 		 and  is more general than , then remove I (this step is
done

		 		 to avoid redundancies in UM).

In step 1) of the procedure, we are looking for the action schemata that have instances present in each active CM. In particular, it is possible that the various CMs in CM-SET contain different actions having a common more general action A in the Generalization Hierarchy. In this case, since A subsumes them, it may be selected for representing the user's intentions at a more abstract level. Clearly, we look for the most specific subsuming action A to establish the user's intentions.

In step 2), we update the UM by adding to it the intention to perform the actions identified in step 1). In particular, in step 2a) we decide how to instantiate the action schemata identified in step 1), according with the restrictions on the parameters present in the active CMs. For example, if there are two active CMs containing the actions USE-PC and USE-WORKST, in step 1) we select the USE-COMPUTERS action (see Figure ). Moreover, since in the two more specific actions the ``comp'' variable is restricted respectively to the PC and Workstation concepts, we restrict it to the Computer concept, that subsumes the previous ones (this is a trivial case, because it is sufficient to restrict the variable to the concept specified in the USE-COMPUTERS action. However, in more complicated cases, other restrictions can be inherited from the CMs, so it is not immediate to identify the most specific concept that is more general than those associated to the parameters of the action in the various CMs).

In steps 2.b) and 2.c) we update the UM with the identified intentions. In particular, in 2.b) we add the new intentions to the UM, while in 2.c) we remove the intentions more general than the new ones that could possibly already have been present in the UM before the application of the procedure, and are at this point redundant. Note that when the action identification produces some candidate actions that may be attached to the active CMs in more than one point, different CMs are created to express the ambiguity. However, they maintain the inner structure of the original CMs, since they only differ with respect to the new connected actions. Moreover, if the new actions introduce additional restrictions on the parameters of the actions already belonging to the CMs, the intentions previously introduced in the UM are still consistent with the new information and it is not necessary to modify them: only the restrictions on their parameters must be updated.