constructing NL interfaces for expert systems.
While interacting with the user the IDEAL system
provides the sufficient linguistic universe for
comprehending, interpreting and translating NL
phrases to a knowledge representation. The
environment is capable of mapping a domain
knowledge to a transformation language, such as,
a Geometric Transformation Language (as used in
CAD systems, Computer Graphics, etc), a Digital
Image Processing Language as in Batchelor (1986),
a Digital Signal Processing Language for digital
filters sinthesis and/or design as in Nie et ali
(1991), etc.
152 Generalities
The solution to a domain problem always involve a
great quantity of knowledge which sometimes is
not well organized to provide the best solution,
That happens in all of the application domains
and so it is not different in the Digital Image
Processing Domain, specially when dealing with
radar or multiespectral satellite images that
have much information, but yet are not well
interpreted. One of the problems arises from the
fact that when using Digital Image Processing
Software, the domain experts dedicate much of the
time to learn how to operate the systems to
choose procedures and provide parameters,
Corr et ali (1989) present a system for automatic
knowledge-based segmentation of remotely sensed
images. Time sequences of remotely-sensed data
information are used together with cartographic
map data and domain expertise in modelling the
scene in terms of segments and their possible
classes. Srinivasan and Richards (1990) use
knowledge-based procedures to provide a new
scheme for incorporating several knowledge
sources in the classification process, Silva and
Bittencourt (1991) considers the use of several
knowledge representation techniques to represent
the knowledge involved in interpretation of
meteorological radar images. The knowledge
Sources are represented using frames, semantic
networks, and/or production rules representation
schemes within a blackboard architecture that
permits different knowledge sources share the
same data, simultaneously or sequentially. In
this system the weather is monitored by the radar
and the images are numerically processed before
the representation, In critical cases the system
activates an alarm calling for experts
interference.
There exist many knowledge-based systems that try
to solve specific Digital Image Processing
problems specially those related to automatic
Image Interpretation. In all of them, however,
the MMI is not efficiently achieved, once the
user (domain expert) is required to participate
through out the whole process. And then, the
tasks are not automatically performed from the
user's point of view. The decisions are always
taken by the user that must be an expert in the
application domain and the systems used.The aim
of this work is to provide an environment through
which the systems may be well used to solve
problems in a specific domain, Using NL to
communicate with the system the user may specify
his intention, avoiding unnecessary computation,
and the tasks are only performed after the
correct procedures and parameters specifications,
Therefore, the tasks are automatically performed
from the user's point of view.
380
The proposed environment translates the user's
query to a tranformation language, that is, the
environment maps the query to the procedures
specification to be adopted for task execution,
The parameters are still provide by the user. But
the environment friendly requests them, sometimes
explaining the use of the parameters and giving
examples, This characterizes a user training
process through the system that can detect any
specification fault before the execution process.
That happens because the domain knowledge is also
represented and so environments like this may be
used in education tasks in remote sensing.
The greatest advantage of using NL in intelligent
environment is that instead of dedicating time to
learn how to specify procedures and parameters to
efficiently use software systems, the user can
dedicate his time searching for new problem
solution estrategies and methods, thus enhancing
the solution space,
2. SYSTEM DEVELOPMENT
2.1 Environment Architecture
This work does not focus on the knowledge
representation of remotely sensed images. This
has been studied by several researchers as
mentioned in section 1,2 and can be adapted in
the case of the digital image processing domain
or any other domain. The objective of the work is
to improve MMI in software usage. That is, turn
parametric and difficult to use systems into
friendly environment, With this kind of system
the user may specify his queries using his own
language within the restricted domain.
The environment developed is composed of modules
for linguistic interpretation, translation to the
set of actions, task execution, and graphic
representation. The linguistic and domain
knowledges are represented using Minsk's frame
representation conception as in Winston (1975),
that permits stereotype representation such as
the description of an object (e.g., a chair) or a
situation (e.g., going into a classroom). These
knowledge sources are used within a blackboard
approach, so that, knowledge sources responsible
for inference rules may share the same data,
searching for the solution to the specified
problem, see figure l.
24.2 Environment-Description
The application domain considered is restricted
to 2D computer graphics where it is possible to
perform some tasks on regular geometric objects
such as rectangles, triangles, squares, circles
and lines, Examples of tasks are translation,
scaling, color alteration and rotation. To
implement the frame representation of this domain
a frame language was used based on the one
proposed by Arariboia (1988),
The prototype was developed in the C language for
graphic presentation and PROLOG for knowledge
representation and inference rules
implementation. Both languages share a knowledge
source where the graphic status is represented.
As the application domain changes the inference
rules and the knowledge necessary may also
chai
whe:
Lan;
the
seci
sem:
one
The
frar
bui:
obj«
est:
or
Afte
lin;
que:
the
con:
are
pro!
con:
Othe
prir
fol:
idei
is,
dom:
loc
kno
att:
size
if;
alt
as,