In order to illustrate the benefits of an integrated development of the UM and
the CMs during plan recognition, we consider an example concerned with the task
of preparing for a CS examination (see Figure 
). We suppose
that students preparing for an examination have to study the topic of the
course (STUDY), do a laboratory work (DO-LAB-WORK) and prepare a report
(PREPARE-REPORT). While the report may be written using the PCs
(USE-PC), the laboratory work generally consists of the development of a
program for solving a simple problem and is written and run on
the workstations (USE-WORKST). Both PCs and workstations are located in the
computer labs of the department. Students may access the computer labs only
if they possess a lab pass that identifies them (see the description in the
introduction of the paper).
We will now examine in detail the dialogue sketched in the introduction, repeating it here for the reader's convenience.
At the beginning of the dialogue, there is no specific information available
about the user and the UM only contains the information belonging to the
Human stereotype: e.g. humans know how to perform the elementary
actions of the domain (Know-act(IS, ASK-FOR), ...).
Figure: Active CM and UM after sentence A1.
In the analysis of sentence A1 the action-identification step of the
plan-recognition algorithm identifies the PREP-CS-EX action, creating an
initial CM; the application
of the UM acquisition rules adds to the UM the information that the user knows
about examinations, and in particular about the DS examination. This is not
sufficient to activate a particular stereotype, like Student, because
other people (e.g. faculties) know very well about CS courses and examinations.
No expansion rules are applied; disambiguation rules are not applied either.
The
upward-expansion step adds the PREP-AND-TAKE-CS-EX action, producing CM1
(see Figure ). The select-common-actions procedure introduces
into the UM the intentions associated to the two identified actions
(Intend1(IS,PREP-CS-EX(IS,DS)), Intend1(IS,PREP-AND-TAKE-CS-EX(IS,DS))). Since
DS is a first year examination, one of the triggering conditions of the
Beginner-student stereotype is satisfied (the user wants to prepare for
a first-year examination) and the UM expansion rules activate the stereotype,
adding to the UM its contents. Being Beginner-student more specific than
Student, the second stereotype is activated as well,
and its contents are added to the UM. The remaining steps of the
plan-recognition algorithm don't modify CM1 and UM, so we do not list them
(in the same way, we will ignore in the rest of the description all the steps
that don't modify the status of the active CMs and the UM). The output of the
analysis of sentence A1 is shown in Figure .
Figure: Active CMs after the analysis of sentence A2.
In the analysis of sentence A2 the action-identification step identifies the GET-LAB-ACCESS action, that has the condition expressed by the user as an effect. The focusing step produces two CMs (CM1' and CM1''), corresponding to the two alternative paths connecting the identified action to the previous CM. The application of the select-common-actions procedure adds the following formulae to the UM:
Intend1(IS, USE-COMPUTERS(IS,comp))
where comp=PCcomp=Workstation
(it subsumes USE-PC and USE-WORKST)
Intend1(IS, GET-LAB-ACCESS(IS))
(common action of CM1' and CM1'')
At this point, the UM suggests that the user is a beginner student, and that
the goal of renewing the application for
the access to the computer lab is not among the user's expected goals.
Since of the two alternative actions
APPLY-TO-LAB-ACCESS and GET-LAB-ACCESS only the first is feasible for the user,
the add-more-specific-actions procedure adds APPLY-TO-LAB-ACCESS to the
active CMs, and the downward-expansion
adds the actions in its decomposition, because it is clear from
the user's question, and from the active stereotypes, that the user does not
know anything about this action. Then, the two CMs are analyzed for
deciding whether the ambiguity between them is relevant or not. Since both of
them involve the same answer to the user's question, the ambiguity is not
relevant and the system selects the contents of the answer to be given, on the
basis of the active CMs. The situation of the CMs
after the analysis of sentence A2
is represented in Figure .
As explained in the introduction, the presence of the user model was necessary for identifying the APPLY-TO-LAB-ACCESS instead of the RENEW-APPL-TO-LAB-ACCESS action: without the analysis of the contents of the UM, the system should have started a clarification dialogue (``Do you already have an expired pass?'') before going on with the analysis and answering the user's question. In fact, the two tasks of applying and renewing the application for laboratory access are different and a relevant ambiguity would have arisen.