The MultimediA DIscourse System (MADIS)
is a dialogue management module developed to provide information
users with answers to single and multiple requests related to
a given design application, using animation, text, graphics,
sound and (eventually) speech recognition. The system tailors
its responses to a particular user based on a user profile and
on the content and sequence of previous exchanges.
MADIS was applied to conceptual database
design (Parent, 1997), to user interface design and more recently,
in collaboration with Protogon Inc. in Montreal, to the design
of ship compartments. The following two images illustrate a sample
knowledge-base created by Protogon for the Automatic Design Deficiency
Constraint Checker, which was a part of the Navy's "Human
Factors Engineering Intelligent CADD" (HFEICADD) project.
The Constraint Checker is an artificial intelligence application
used to identify design flaws based on CAD (Computer Aided Design)
drawings of ship compartments. The design errors found by the
Constraint Checker are shown by "glyphs" - interconnected
red squares placed on top of the CAD objects that cause problems.
The "3D simulation" is the model of the compartment.
MADIS was developed in JAVA. In addition
to platform portability, it allows a system to be run remotely
by end-users visiting a web site. Various intelligent help applications
may be developed with the dialogue module. MADIS can provide
interactive help from an Internet browser running concurrently
with another software application containing needed information.
Utterances
The utterances MADIS can address include
requests and assertions. They are:
Requests
(1)What is the definition of... ?; (2)
Tell me about... ; (3) Should I...? ; (4) Should I...or...?;
(5) How do I decide on ...? (6) How do I decide if I should ...?;
(7) How do I decide if I should...or...? (8) At what point should
I ...? and (9) Clarify?
Assertions
(1) Offer (information); (2) Assert;
(3) Promise; (4) Express content; (5) Express discontent; (6)
Withdraw request; (7) Reject offer; (8) Withdraw offer and (9)
Reject request.
Overview of Usability Issues
The following list of issues, considered
important to the success of human communication was assumed desirable
of an ideal discourse management system. Research goals were
therefore elaborated based on this list. Usability issues are
presented in decreasing order of importance.
1) Ease of learning: The user should
be able to learn how to send and receive messages within a few
minutes. This procedure should be transparent.
2) Accessibility: The discourse facility
should be available at all times.
3) Coherence: There should be inferable
links between the ideas and topics in successive discourse segments
(e.g. within a message or a discourse episode). More specifically
each utterance within a message should refer to at least one
topic discussed in the previous utterance. Each message should
also have a logical link to the user's request or comment.
4) Inference of intentions: The intentions
behind the system's utterances should be clear to the user. For
example, when MADIS presents the user with a message, it should
be clear to the user that the system is responding to a question
or an assumed need.
5) Acceptability: The system should
infer the user's goals and use these in tailoring its communications.
For example, when the user asks "How do I decide on the
colour of my screen"? and then requests the definition of
a word contained in the system's response, the system can infer
that its message was unclear and reformulate it using the definition
of the word requested.
6) Information completeness: The system's
messages should be tailored to the user's previous knowledge
(e.g. experience with the system). The information provided should
not be incomplete nor excessive. The right amount of information
should be given at least 7 out of 10 times.
7) Context: MADIS should take into account
aspects of the discourse context such as previous requests. Context
can be derived from multiple exchanges within a single dialogue
episode or between episodes. For example, if the user asks the
system to clarify a message twice, the system can ask for specification
of that which is unclear (e.g. the system's intention, the vocabulary
or sentence structure).
8) Interactivity: The system should
allow interruptions such as the cancellation of a request.
9) Cohesion: Surface level ties should
be made between textual elements (e.g. use of pronouns).
10) Speed: The system should respond
to the user within 2 seconds of his/her selection or give some
indication that it is processing the information.
11) Correctness: The system behaviour
should follow the functional specifications 90% of the time.
12) Robustness: The system should not
fail completely when the normal operating conditions are not
met. Instead when faced with ambiguity, the system should advise
the user of its need for clarification. For example, ask the
user to repeat all or part of the request.
Related Publications
Brahan, J.W., Farley,B., Orchard, R.A.
Parent, A. Phan, C.S. (1992). A designer's consultant, Proceedings
of Expert Systems '92, the Twelfth Annual Technical Conference
of the British Computer Society Specialist Group on Expert Systems,
Cambridge, U.K. December 15-17. NRC C3234.
Boulet, M.-M., Brahan, J.W., Cole, A.J.,
Duchastel, P., Hartley, J.R., Orchard, R.A., Parent, A., Pilkington,
R., Phan, C.S. (1989). A design task advisor . Artificial Intelligence
and Education. Proceedings of the International Conference
on AI and Education, Amsterdam, The Netherlands. May 24-26,
Published by D. Biermans, J. Breuker, and J. Dandberg. pp. 25-31.
NRC30444.
Boulet, M.M., Pascot, D., Parent, A.,
Slobodrian, S. (1988). Acquisition de connaissances pour un système
conseiller méthodologique, Informatique cognitive des
organisations (ICO), novembre, NRC 31126
Parent, A. (1990). Collection and analysis
of dialogue protocols for the question-answering facility of
the entity-relationship modeling advisor. Technical Report
ERB-1032, October. NRC 31827
Parent, A. (1991). Analysis of dialogue
protocols. Proceedings of the 4th University of New Brunswick
Artificial Intelligence Symposium, Fredericton, N.B., 19-21
September. NRC 33149.
Parent A. (1993). Un module question-réponse
pour la conception, Intelligence Artificielle au Canada,
hiver, NRC 35022.
Parent A. (1997). Analysing design-oriented
dialogues: a case study in conceptual data modelling, Design
Studies, Vol 18, pp. 43-66. NRC 40153.
Parent, A., Farley, B. (1994). Un module
question-réponse pour la conception: définition,
formalisation et implantation. Informatique cognitive des
organisations (ICO), automne 1994, 6(3), pp. 59-67. NRC 38353.
