provides the necessary parameters ( X,y,w,z). In
this process he does not need to know much of the
graphic system he is using.
In fact the primitives are acting on the
represented knowledge using the frame language
referred to previously. This makes the user's
procedures and parameters specification a high
level abstraction task,
As mentioned before the knowledge representation
and the inference rules are specified using the
PROLOG language. The PROLOG language is a
declarative language that permits easy rules
coding for users, Besides it has an inference
motor adequate for the proposed environment.
The environment can be adapted to any other
domain as mentioned previously, Other inference
rules may be needed and then the user must code
other primitives to perform the tasks, In the
Digital Image Processing domain the environment
may be able to translate queries as in example 2
(interactions in Portuguese).
Exemplo 2:
USUARIO : Suavizar a imagem.
(USER : Smooth image)
AMBIENTE : Especifique a imagem a ser
filtrada.
( ENVIRONMENT : Specify the image to be
filtered)
USUARIO : (Fornece o nome da imagem)
( USER: (Provides the name )
AMBIENTE : O filtro passa-baixa deve
preservar a borda?
(ENVIRONMENT : Is it an edge preserving
smoothing filter?)
USUARIO : (0 usuario pode responder de
acordo com a pergunta do sistema ou pode
indagar o que significa preservar a borda, ao
que o sistema pode atender explicando o
efeito de preservarcao da borda e
exemplificando)
(USER : May answer accordingly or ask about
edge preserving which the environment may
explain and give examples).
All the information provided in example 2 may be
Stored so that in the next user's interaction
requesting the same processing the environment
can promptly perform its tasks recalling the
definitions stored.
2.3 Results Obtained
The environment discussed here is implemented for
the 2D Computer Graphics domain. Instead of
trying to solve all of the problems concerned to
2D transformations the work aims to show the
usefulness of designing Natural Language
environments for Man-Machine Interaction in
system development in any application domain. The
task involves people dealing with Artificial
382
Intelligence such as Knowledge Engineers, and
domain experts such as those working in Computer
Graphics, Remote Sensing, Image Interpretation,
Digital Signal Processing, etc.
The way the prototype is designed permits some
interactions with the user, So that environment
successfully performs tasks as stated in example
le While building up his request the user
maintains a high level dialog with the machine.
That is, he specifies everything in his own
language and does not get involved with detailed
parameters specification, Thus establishing an
improved Man-Machine Interaction.
Providing the linguistic universe for Natural
Language interaction the environment can show its
efficiency with people without much experience in
computer use and/or the application domain
considered, The storage of knowledge involved in
problem solution permits that most of the
situations be treated automatically from the
user's viewpoint thus avoiding redefinition of
the user's intention,
3, CONCLUSIONS
The prototype developed showed the possibility of
using Natural Language in knowledge-based systems
which improves Man-Machine Interaction. Thus
users can face computational systems as real
helping tools once they do not need to get
involved with detailed specifications that make
them learn several lessons on systems use before
efficiently use them to solve their problems.
Then time used before for training purposes can
now be dedicated to new solution strategies and
methods discoveries. Besides, the tasks can be
executed automatically under user's viewpoint, so
that, the efforts for problem solution can be
measured and evaluated.
Natural Language has not been used in system
design because of the problems involved, It is a
very dynamic subject where the rules are
constantly changing according to customs,
culture, etc., within a region. This means new
rules and data storage. Also, the same feeling
can be expressed in several different ways which
requires a complete linguistic knowledge
representation that may require great memory
capacity from the hardware used, Restricting the
domain it is possible to use free NL to treat a
reduced number of situations,
The initial work objective was to design a system
where the Remote Sensing experts would perform
several Digital Image Processing activities,
specially those related to Image Interpretation
involving different knowledge sources. The
greatest problem found concerned image knowledge
representation, And then, an environment for 2D
Computer Graphics domain was developed. The
objects treated in this domain are simple to be
represented, However, image knowledge
representation has been studied by some authors
already and the prototype described can now be
tested in Image Processing. The same system
philosophy design has been used to develop an
interface for an image processing software that
performs morphological processing on images.
Vis
Env
198
Pra!
Ltd.
Pro
Vis:
BOR!
Ides
Lin
PP.
Idei
Ana:
Ling
pP.
Mes
Est:
Lin;
PP.
CORI
Ces
Pro;
vol.
Rep:
the
920.
HAYI
Cont
July
NIE,
Peo
proc
filt
2, À
NII,
Mode
Blac
198¢
OLI!
Nati
OLI!
Tese
515]
PASS
Espe