The Development of a Bilingual Fuzzy Expert System Shell


J. King Saud University, Vol. 21, Comp. & Info. Sci,  pp. 27-43, Riyadh (2009/1430H.) 

 

The Development of a Bilingual Fuzzy Expert System Shell 
 
 

Hassan Mathkour, Israa Al-Turaiki and Ameur Touir  
 Department of Computer Science 

 King Saud University 
 Riyadh, Saudi Arabia 

mathkour@ccis.ksu.edu.sa, binmathkour@yahoo.com, touir@ksu.edu.sa 
 
 

(Received 21/12/2008; accepted for publication 02/03/2009) 
 

 
Abstract. Fuzzy logic has been incorporated in many expert systems to solve real world problems that are 
inherently ambiguous. With fuzzy logic it is possible to program human intuition through the development of 
fuzzy expert system shells. A fuzzy expert system shell is a tool that helps build expert systems to manage fuzzy 
problems. Commercial as well as non-commercial fuzzy expert system shells are available. These shells provide 
variety of functions to facilitate the development of fuzzy expert systems for real world problems in different 
application areas such as medicine, engineering, and finance. To the best of our knowledge, none of the available 
fuzzy shells is natively developed for the Arabic language. This paper describes the development and the 
experimentation of a bilingual fuzzy expert system shell. This shell is intended to be a research tool for fuzzy 
expert systems developers in bilingual environments similar to those in the Arab world where users and developers 
use multi-languages due to their educational backgrounds and working environments. The shell processes fuzzy 
terms of the Arabic language as well as the English language. The shell is a general purpose shell that provides 
users with the ability to develop Arabic/English fuzzy expert systems using a simple Graphical User Interface. It 
applies implication methods that bear resemblance to human intuition. In the process of the development, a 
comparison of various fuzzy expert system shells has been performed to identify strengths and weaknesses of 
available shells. Experiments with our shell are reported and its performance is compared to existing shells that 
use different implication methods. 
 
Keywords:  Expert system shells, knowledge based systems, Fuzzy Implication methods, Bilingual Systems. 

 
1. Introduction 

 
Expert system shells are versatile tools that are used 
to create expert systems.  Fuzzy expert system shells 
have been developed to allow for reasoning that 
deals with crisp and fuzzy sets. These shells allow 
incorporation and manipulation of imprecise 
information using fuzzy set theory developed by 
Zadeh (Zadeh, 1965). They are used to create expert 
systems that can handle imprecise situations 
effectively. The ability to operate under imprecise 
environment makes expert systems closely behave 
like human being and provides a natural 
representation of people's daily terminologies. The 
ability of treating ambiguities, in a manner similar to 
human experts, makes expert systems versatile and 
adaptable to unforeseen circumstances which are 
difficult to avoid in real life applications. This has 
made fuzzy logic a suitable means to deal with the 
fuzziness of data and knowledge frequently 
encountered in the terminologies of human experts 
when developing knowledge based systems (Kelmet 
and Slany, 1993).  
 

There have been attempts to design fuzzy expert 
system shells for large-scale general-purpose as well 
as domain specific applications (Philip, 1991; Aly 
and Vrana, 2006). Over the years, a large number of 
expert system shells have been developed and 
several of them are commercially available. JFK 
(López-Ortega, 2006), FuzzyShell (Pan, 1996), 
FuzzyJess (Orchard, 2001), FuzzyCLIPS (Orchard, 
2004), FLINT (Shalfield, 2005), FLOPS (Siler and 
Buckley, 2005), Fuzzy Logic (Mathworks, 1999), 
and FuzzyJ toolkit (Council, 2001; Orchard, 2001) 
are examples of expert system shells. We have 
analyzed several of the existing shells in an attempt 
to indentify a shell having features that natively 
supports application development in Arabic language 
while allowing for application development in other 
languages. We searched for a shell that 
accommodates for Arabic fuzzy terms naturally and 
which employ intuitional inference methods. Our 
unsuccessful endeavor and realizing that making 
such a shell available will be useful for bilingual 
developers and users in research and educational 
environments motivated us to design and implement 
a fuzzy shell with Arabic/English support.   

27



28 Mathkour et al.: The Development of a Bilingual Fuzzy Expert System Shell 
 

 

In the process, we have found it helpful to furnish 
a comparison for a set of the available fuzzy shells.  
These shells differ from one another in several 
aspects.  For example, most of the shells implement 
inference methods that are mentioned in (Zadeh, 
1975; Mamdani, 1977) while many (Fukami, 1980; 
Mizumoto, 1981; Mizumoto, 1982) have advocated 
that the methods that are based on the interpretation 
given in (Mizumoto et al., 1979; Mizumoto et al., 
1979; Mizumoto et al., 1979) perform better as they 
induce human intuition. In this research, we have 
taken the interpretation that is supported in 
(Mizumoto et al., 1979; Mizumoto et al., 1979; 
Mizumoto et al., 1979; Mizumoto, 1981). Our work 
in (Mathkour et al., 2009) introduces an Arabized 
fuzzy expert system shell. In this paper, we present 
the development of a bilingual (Arabic/English) 
fuzzy expert system shell, which is an extension of 
our work in (Mathkour et al., 2009), to allow for 
both the Arabic and English languages. We also 
report on experiments with the shell using real life 
data to demonstrate and analyze its human-like 
behavior using the selected inference methods. To 
measure its effectiveness, we have compared its 
performance with some of the available shells. We 
report on the experiments and comparison of our 
shell with FuzzyClips (Orchard, 1996) and FuzzyJ 
(Council, 2001; Orchard, 2001). 

 
The objective of our extended shell is to provide 

a comprehensive tool that is intended to be a 
research tool for fuzzy expert systems developers in 
multi-lingual environments similar to those in the 
Arab world where users and developers use multi-
languages due to their educational backgrounds and 
working environments. It is a general purpose shell 
that is based on the implication methods: Rs, Rg, Rgs, 
Rgg, Rsg and Rss (Fukami, 1980; Siler and Buckley, 
2005; Mizumoto et al., 1979; Mizumoto et al., 1979; 
Mizumoto, 1981). 

 
It is also observed that many shells use dedicated 

programming languages for the expert system 
application development. Consequently, application 
developers are required to learn the programming 
languages that are supported by these shells. 
Learning a new programming language is not a 
desired requisite, especially for those who do not 
have a programming aptitude. Learning a new 
programming language distracts developers from 
their main objective of developing expert systems in 

their specific domains. In our shell, we have used a 
visual environment by adopting a simple graphical 
user interface. The interface supports both Arabic as 
well as English languages and it can be tailored for 
other languages by adding the user interface support 
for the required language.  
   

In (Mathkour et al., 2009), we developed 
comparison criteria to evaluate aspects of available 
expert system shells. The criteria include evaluation 
of end-user interface, developer Interface and 
availability and installation of shell. In this paper, we 
further discuss these criteria and employ them to 
formulate comparison tables of a larger number of 
existing expert system shells. 

 
The rest of the paper is organized as follows: 

Section 2 presents a comparison of twenty expert 
system shells along with brief description of the 
comparison criteria. Section 3 presents the developed 
fuzzy shell, describes the implication methods, and 
the implementation. Section 4 presents 
experimentation with the system. Section 5 presents 
a comparison of our shell to some existing ones. 
Section 6 concludes the paper. 

 
2. Comparison of Existing Shells 

 
We have endeavored to compare the features of 

twenty shells of those available commercially and 
otherwise. These include Fuzzy Logic(Mathworks, 
1999), JFK (López-Ortega, 2006), FuzzyJess 
(Orchard, 2001), FuzzyCLIPS (Orchard, 2004), 
FuzzyShell (Pan, 1996), FLINT (Shalfield, 2005), 
FLOPS (Siler and Buckley, 2005), CLIPS 
(Giarratano, 1998), Jess (Friedmann-Hill, 1999), 
Flex(Vasey, 1996), PSS (Forgy, 1981), ESB (Kent 
and Denholm, 1990), ESBuilder (Ishihara et al, 
1995), and FuzzyJ toolkit (Council, 2001; Orchard, 
2001). First we present a discussion of the 
comparison criteria, then present the results of our 
comparison in Table 8. 
 
2.1 End-user interface 

The user interface is an important component of 
any software development tool as it allows 
interaction between application developers and the 
tool. The user interface must be natural in the context 
applications that are being developed thereby 
releasing the developers from learning extraneous 
concepts and focusing on the development issues.  



 J. King Saud University, Vol. 21, Comp. & Info. Sci, Riyadh (2009/1430H.)          29 

 

The quality of the user interface is judged by its ease 
of use and naturalness. The following features are 
indicators of the quality of an interface: 
1. Explanation facilities: This is used to explain 

the process through which the system has 
arrived at a decision.  

2. User friendliness: This is judged by the quality 
of graphical user interface components such as 
menus, buttons, and usage of a natural 
language. 

3. The ability to change the earlier answers 
without having to repeat the session from the 
beginning. 

 
2.2 Developer interface 

The expert system developers enter their 
knowledge through the rule editor. The rule editor 
should support the rule type selection and creation, 
rule change and update process, mathematical 
operations to implement the inference engine 
strategies, built-in member functions, de-
fuzzification methods, certainty factor handling, 
error correction, and fact refinement and 
documentation. In addition to these, the rule editor 
must have provisions to interact with external 
environments like DBMSs, Spread sheets and 
Programming in modern languages like Java and C#. 
Features related to the rule editor are shown in Table 
1 to Table 5 with their respective weights. 
 
2.3 Procurement and installation 

The availability of these tools could be 
problematic in some linguistic regions of the world. 
Once available, their installation is not always 
straight forward. Hence we have used it as an 
evaluation factor. Table 6 and 7 shows the weight 
assigned to measure the ease of procurement and 
installation.  
 
Table 1. Rule type weight 

Rule Type Weight 
Complex IF-THEN-ELSE rule 5 
Complex IF-THEN rule 
(multiple antecedents or/ and  
multiple consequents) 

3 

Simple IF-THEN rule (one  
antecedent one consequent) 

1 

 
 
 

 
Table 2. Rule chaining weight 

Method  Symbol 
Forward F 
Backward B 
No built in chaining strategy 
(user defined) 

NA 

Table 3. Math capability  weight 
Supported Math Functions Weight 
Advanced math functions  5 
Basic math functions  3 
None 1 

Table 4. Inference strategy weight 
Supported Inference Strategies Weight 
None 1 
One or Two  3 
Three 5  

Table 5. Documentation weight 
Documentation  Weight 
Comprehensive & easy to read 5 
Brief  1 

Table 6. Procurement weight 
Procurement Method Weight 

Download from the Internet 5 

Order package CD  1 

Table 7. Installation weight 
Installation  Method Weight 
Unpack (run) one file 5 
Unpack source and compile 1 

 
3. The Proposed Bilingual Fuzzy Expert System 

Shell 
The entry point to the system provides the users 

with the option of building expert systems using 
Arabic or English knowledge bases (Fig. 1). Upon 
selection, Arabic or an English screen portraying the 
main components of system is displayed (Fig. 2.a 
and 2.b). The main components of the shell are the 
variable editor, rule editor, and the inference engine. 
 
3.1 The variable editor 

The variable editor’s main purpose is to provide 
functions to create, edit, and delete fuzzy variables, 
their fuzzy values, membership functions, and 
universe of discourse. The layout of our variable 
editor is shown in Fig. 3.a and 3.b. The variable 
editor can be launched from the menu button of 
“Variable Editor”  “محرر المتغيرات” in Fig. 3. Created 
variables and their properties can be seen from a 
dropdown menu. 



30 Mathkour et al.: The Development of a Bilingual Fuzzy Expert System Shell 
 

 

 
 
 
 

; 

a - - • - 1-' - - - - - - - 1- - - -
"a - - . -1- • - - - - - - - 1- - - -
~ - - - - - - - - - - - - - 1- - - -
~ - - - - - - - - - - - - - 1- - - -
:- ........ - - 1- ;; .... - ~ _.... - - I........ - ..... 
~- - - - - 1- - - - - - - - - 1- - - -
.~ - - - - -1- ' - - - - - - - 1- - - -

.! ......... - - I ..... ;; - - - -.... .... ..... I....... .... ..... 
J - - - - 1- ;; - - - - - - - 1- - - -

~... ..... - - - I ..... ~ ...... - - - .... - I......... .... .... 
•• - - - - - - - - - - - - - 1- - - -

J! --:Ii - - ..... - I..... ... ... ... ... ... - ..... - I.... - .... .... 
0- - - - - 1- - - - - - - - - 1- - - -
~ - - - - 1- - - - - - - - 1- - - -
~" - - - - 1- - - - - - - - - 1- - - -

~.ti - - - -1- - - - - - - - - 1- - - -
~ - - - - 1- · - - - - - - - 1- - - -
~ - - - - 1-0 1- - - - - - - 1- - - -
LR - - - - !:..:L - - - - - - = != : : : 
~ - - - ..... I-:.if-....... ..... .......... ..... 1-' 

~. ~ tt· ~ !!j~hJ!,,-=i Hi ---'=-----1 
"'''''' ....... " _ _ '<I.O;i 



 J. King Saud University, Vol. 21, Comp. & Info. Sci, Riyadh (2009/1430H.)          31 

 

   
Fig. 1. System entry point. 

 

   
(a) 

  
(b) 

Fig. 2. The arabic and english components of the 
system. 

 

3.2 Rules and Rule-base Editor 
For The rule-base editor permits application 

developers to create, edit and delete rules. The rules 
are of the form IF antecedents Then consequents. A 
rule may have more than one antecedent and one 
consequent. Also, the editor allows the “Else part” in 
the rules. We use rules as our knowledge 
representation scheme because they are natural in 

representing expert knowledge, and they are easier to 
understand, modify, and maintain. 
The rule editor can be launched from the menu 
entitled “KB Editor”  “ المعرفةمحرر قواعد  ” on the menu 
bar of Fig. 2. Figure 4 shows the layout of the 
Knowledge base editor at the creation of a new 
knowledge base. A new knowledge base is created 
using “New KB” “ادخال قاعدة معرفة “ in the menu bar 
of Fig. 2. 
 

 
(a) 

 
(b) 

Fig. 3. Arabic//english variable editor layout. 
 

 
(a) 



32 Mathkour et al.: The Development of a Bilingual Fuzzy Expert System Shell 
 

 

 
(b) 

Fig. 4. The arabic/english layout of the knowledge 
base editor. 

 
3.3 Inference rngine 

The inference engine uses implication, 
composition, aggregation and linkage as given in 
(Leung and  Lam, 1988; Aly and Vrana, 2006; 
Bandler and Kohout, 1980), and briefly described in 
the following subsections. It is the part of the 
knowledge based system that is responsible for 
deriving conclusions from existing data, i.e., deriving 
new knowledge from existing ones. 

 
3.3.1 The implication methods 
Our shell has a backward inference engine and uses 
the implication methods discussed in  (Fukami et al. 
1980; Mizumoto et al., 1979; Mizumoto et al., 1979; 
Mizumoto et al., 1979, Mizumoto, 1981), namely, 
Rs, Rg, Rgs, Rgg,Rsg, and Rss. Details of the 
inference methods are found in (Mizumoto et al., 
1979; Mizumoto et al., 1979). The choice of the 
inference methods is based on the observation that 
such methods closely mirror the human intuitions as 
compared to those in (Zadeh, 1999; Zadeh, 2006; 
Zadeh, 1975; Mamdani, 1977). This has been 
advocated in previous work (Mizumoto, 1981; 
Mizumoto, 1982). The shell allows the user to either 
use all the implication methods or select one of them. 
Figure 5 shows a screen shot of the working of the 
system when conclusion is obtained using the Rs 
Implication method. The conflict resolution strategy 
used in our shell is the most specific strategy. 

 
A fuzzy inference method needs to satisfy the 

criteria shown in Table 9, in order to resemble 
human intuition (Fukami et al, 1980; Mizumoto, 
1981; Mizumoto, 1982). The inference methods 
presented in (Zadeh, 1994) do not satisfy the criteria 
in Table 9, except CriterionIV-1. The inference 

methods in (Mamdani, 1977) on the other hand 
satisfy Criterion I and II-2.  Criterion II-2 is 
applicable when there is no strong relation between 
“x is A” and “y is B”. In criterion IV-1, information 
about y cannot be inferred from the conditional 
inference “if  x is A then y is B” when “x is not A”.  
Details of related issues are found in (Bandler and. 
Kohout, 1980; Willmott, 1980; Mizumoto, 1981, 
Mizumoto, 1982).  

 

 
(a) 

 
(b) 

Fig. 5. Conclusion obtained using Rs implication. 
 

Following the criteria in Table 9, fuzzy 
inferences can be classified into the following four 
types. Illustration of criteria that are satisfied by the 
implication methods Rs, Rg, Rgs, Rgg, Rsg, and Rss 
is given in Table 10 (Fukami, 1980; Mizumoto, 
1981; Mizumoto, 1982). 

 
 Type 1: The binary relation between the 

antecedent A and the consequence B is 
translated into ),( BARs . In type 1 inference, 
Criteria I, II-1, III, IV-1 are satisfied. 

 
 



 J. King Saud University, Vol. 21, Comp. & Info. Sci, Riyadh (2009/1430H.)          33 

 

Table 9. Fuzzy inference criteria 
 
 

Criterion I 

Ant 1: if x is A then y is B 
An t2: x is A 

 
Cons: y is B 

 
 

Criterion II-1 
 

Ant 1: if x is A then y is B 
Ant 2: x is very A 

 
Cons: y is very B 

 
 

Criterion II-2 
 

Ant 1: if x is A then y is B 
Ant 2: x is very A 

 
Cons: y is B 

 
 

Criterion III 
 

Ant 1: if x is A then y is B 
Ant 2: x is more or less A 

 
Cons: y is more or less B 

 
 

Criterion IV-1 
 

Ant 1: if x is A then y is B 
Ant 2: x is not A 

 
Cons: y is unknown 

 
 

Criterion IV-2 
 

Ant 1: if x is A then y is B 
Ant 2: x is not A 
 
Cons: y is not B 

 
 Type 2: The binary relation between the 

antecedent A and the consequence B is 
translated into ),( BARg . In type 2 inference, 
Criteria I, II-2, III, IV-1 are satisfied. 

 Type 3: The binary relation between the 
antecedent A and the consequence B is 
translated into ),( BARsg . In type 3 inference, 
Criteria I, II-1, III, IV-2 are satisfied. 

 Type 4: The binary relation between the 
antecedent A and the consequence B is 
translated into ),( BARgg . In type 4 inference, 
Criteria I, II-2, III, IV-2 are satisfied. 
 

An Rs implication example: 
For the rule, If x is small then y is middle, where 
U=V= 0+1+2+3+4+5+7+8+9+10, 
A=small= 1/0+0.8/1+0.6/2+0.4/3+0.2/4, and 
B= middle 
=0.2/2+0.4/3+0.8/4+1/5+0.8/6+0.4/7+0.2/8,  
Rs(A,B) is given in the Table 11. 
 
3.3.2 Composition 
A fuzzy composition relation R(A,B) of R1 and R2 
is simply the relation obtained by applying R1 and 
R2 one after the other. The most frequently used 

  Table 10. Criteria satisfied by each implication method 
Ant 2 Cons Rs Rg Rgs Rgg Rsg Rss 

A B + + + + + + 
Very A Very B + - - - + + 
Very A B - + + + - - 

More or less A More or less B + + + + + + 
Not A Not B - - + + + + 

 
      Table 11. Rs(A,B) 

 
 
 
 
 
U 
 

V 
 0 1 2 3 4 5 6 7 8 9 10 
0 0 0 0 0 0 1 0 0 0 0 0 
1 0 0 0 0 1 1 1 0 0 0 0 
2 0 0 0 0 1 1 1 0 0 0 0 
3 0 0 0 1 1 1 1 1 0 0 0 
4 0 0 1 1 1 1 1 1 1 0 0 
5 1 1 1 1 1 1 1 1 1 1 1 
6 1 1 1 1 1 1 1 1 1 1 1 
7 1 1 1 1 1 1 1 1 1 1 1 
8 1 1 1 1 1 1 1 1 1 1 1 
9 1 1 1 1 1 1 1 1 1 1 1 
10 1 1 1 1 1 1 1 1 1 1 1 

 



34 Mathkour et al.: The Development of a Bilingual Fuzzy Expert System Shell 
 

 

composition operator in fuzzy logic is the Max-Min 
composition operator in (Zadeh, 1999; Zadeh, 1975) 
and it is the one we used in our shell. 

Max-Min Composition 
Let R be a fuzzy relation in X × Y, and S be a 

fuzzy relation in Y × Z. The Max-Min 
composition of R and S, RoS, is a fuzzy 
relation in X × Z such that 

RoS↔μRoS(x,z) =   v {μR(x,y) ^ μS(y,z) } 
 
A Max-Min composition example: 

Suppose we have the following two relations R 
and S: 

 
 
 

6.00
11.0
8.05.0

1.019.0
06.04.0

SR  

 
max{min(0.4,0.5), min(0.6, 0.1), min(0, 0)} = max{ 0.4, 0.1, 0} = 0.4 
max{min(0.4,0.8), min(0.6, 1), min(0, 0.6)}= max{ 0.4, 0.6, 0} = 0.6 
max{min(0.9,0.5), min(1, 0.1), min(0.1, 0)}= max{ 0.5, 0.1, 0} = 0.5 
max{min(0.9,0.8), min(1, 1), min(0.1, 0.6)}= max{ 0.8, 1, 0.1} = 1 
 
3.3.3The Aggregation and link operators 
An aggregation operator is needed when a rule has k 
conditions. The rule is decomposed into k 
implications. Each implication is used separately to 
infer a value by applying the fuzzy implication The 
values are then aggregated using the aggregation 
operators used in the rule including OR and AND. 
The final result is obtained after a MAX operation 
over the corresponding values inferred by all the 
rules or fuzzy membership functions (Zadeh, 1975; 
Fukami, 1980; Mizumoto, 1981; Mizumoto1982; 
Zadeh, 1999). 
 
3.4 Implementation Issues 

Similar to that in (Mathkour et al., 2009), the 
main data structures used in the implementation of 
the shell are arrays. Since all the implication 
methods used here depend on the Rs and Rg 
operations, there was a need to implement Rs and Rg 
operations as separate methods. Both methods accept 
two parameters and return the result after performing 

Rs or Rg implication.  Each of the six implications 
was implemented as methods that accept two 
matrices and return the result after performing the 
implication of the whole matrices. 

 
4. Experimental Results 

 
The purpose of this section is to illustrate that our 

fuzzy expert system shell produces correct results 
and the implication methods used satisfy the criteria 
mentioned in Table 9. The knowledge base used in 
this experiment is taken form (Ganoud, et al. 2005). 
In (Ganoud, et al. 2005), the authors study the 
influence of random factors on the planning of 
building works. Deciding the exact period of 
building projects is a very difficult job due to the fact 
that building projects are affected by different 
unpredictable factors. The random factors which are 
studied include three factors: the cessation of 
machines, the absence of some professionals, and the 
influence of weather condition. The rules are given 
in Table 12 and the membership functions are given 
in Figures 6 to 9 (Ganoud, et al. 2005). 

 
Fig. 6. The membership function of "absence of 

professionals". 
 
 

 
Fig. 7. The membership function of "weather 

changes". 

 
Fig. 8. The membership function of machine 

cessation". 

R y1 y2 y3 
x1 0.4 0.6 0 
x2 0.9 1 0.1 

S z1 z2 
y1 0.5 0.8 
y2 0.1 1 
Y3 0 0.6 



 J. King Saud University, Vol. 21, Comp. & Info. Sci, Riyadh (2009/1430H.)          35 

 

 

 
Fig. 9. The membership function of the conclusion 

"increase in the period of building". 
 

 
The knowledge base consists of 27 fuzzy rules. All 
the rules have the same number of antecedents. The 
same fuzzy variables are used in all the 27 rules as 
well as the same conclusion (target). The shell was 
run several times, each time with different 
observation. The observations are: 
 
 

 
(a) 

 
(b) 

Fig. 10. Result of observation 1. 
 

 Observation 1: X is low, Y is high and Z is 
high. 

It is observed (as illustrated in Figure 10) that 
using Rg,Rgs and Rgg has given the expected 
results according to the Table 9. On the other 
hand, using the implications Rs, Rsg, and Rss we 
found out upon examining the resulting matrices 
that the results were not exact as expected but 
were very close to what was expected. 

 
 

 
Table 12. The rules data used in the experiment (Ganoud, et 

al. 2005)    
 

 القاعدة

نسبة التعطل 
 الناجم عن
تعطل 
 اآلليات

نسبة التعطل 
الناجم عن 
 غياب العمال

نسبة التعطل 
الناجم عن 
الظروف 
 الجوية

 النتيجة

 عالية عالية عالية منخفضة ١
 عالية عالية عالية متوسطة ٢
 عالية عالية عالية عالية ٣
 متوسطة عالية متوسطة منخفضة ٤
 متوسطة متوسطة متوسطة منخفضة ٥
منخفضة  منخفضة منخفضة منخفضة ٦

ً  متوسطة متوسطة متوسطة متوسطة ٧
 منخفضة منخفضة منخفضة متوسطة ٨
 متوسطة عالية منخفضة متوسطة ٩
 عالية عالية متوسطة متوسطة ١٠
 عالية متوسطة عالية متوسطة ١١
 متوسطة متوسطة منخفضة متوسطة ١٢
 متوسطة منخفضة عالية متوسطة ١٣
 منخفضة منخفضة متوسطة متوسطة ١٤
منخفضة  متوسطة منخفضة منخفضة ١٥

ً  متوسطة عالية منخفضة عالية ١٦
 منخفضة منخفضة متوسطة منخفضة ١٧
منخفضة  متوسطة منخفضة منخفضة ١٨

ً  عالية متوسطة عالية عالية ١٩
 متوسطة متوسطة عالية منخفضة ٢٠
 متوسطة منخفضة عالية منخفضة ٢١
 عالية منخفضة عالية عالية ٢٢
 عالية عالية متوسطة عالية ٢٣
 متوسطة عالية منخفضة عالية ٢٤
 متوسطة منخفضة متوسطة عالية ٢٥
 متوسطة متوسطة متوسطة عالية ٢٦
 منخفضة منخفضة منخفضة عالية ٢٧

 
 

 
(a) 



36 Mathkour et al.: The Development of a Bilingual Fuzzy Expert System Shell 
 

 

 
(b) 

Fig. 11. Result of observation 2. 
 

 Observation 2: X is not low, Y is not high 
and Z is not high. 

It is observed (as illustrated in Figure 11)  that all the 
implications has given  the expected results 
according to the Table 9 except for Rgs and Rss. 
Examining the matrices that resulted from using Rgs 
and Rss we have found that they were very much 
close to the matrix of the expected result which is 
"not high". The result in words should have been 
“more or less not high", but because the shell is not 
designed to handle composite hedges the result 
matrix was translated to "more or less low" 

 Observation 3: X is very low, Y is very 
high and Z is very high. 

It is observed (as illustrated in Figure 12) that the 
implications Rg, Rgs and Rgg have given the 
expected results according to Table 9. On the other 
hand, the implications Rs ,Rsg and Rss have given 
different results than expected.  Upon examining the 
matrices that resulted from using Rs, Rsg and Rss we 
found that they were very much close to the matrix 
of the expected result  which is "very high" . The 
result in words should have been " more or less very 
high" ,but because the shell is not designed to handle 
composite hedges the result matrix was translated to 
"more or less high". 

 
(a) 

 
(b) 

Fig. 12. Result of observation 3. 
 
 Observation 4: X is more or less low, Y is 

more or less high and Z is more or less 
high. 

It is observed (as illustrated in Figure 13) that 
the implications Rs, Rsg, and Rss all have given 
the expected results according to Table 9. But 
for the rest of the implication methods the 
results are close to the expected. 

 

 
(a) 

 
(b) 

Fig. 13. Result of observation 4. 
 

5. Comparison with Existing Tools 
 

We demonstrate that the proposed fuzzy shell 
performs in a natural manner that mirrors the human 
inferences of real world problems and yields the 
expected conclusions that conform to human 
possible conclusions as compared to other tools. The 
comparison is made with tools that use the inference 
methods in (Mamdani, 1977) which are commonly 



 J. King Saud University, Vol. 21, Comp. & Info. Sci, Riyadh (2009/1430H.)          37 

 

used in commercial fuzzy expert system shells such 
as FuzzyClips, and with tools such as FuzzyJ Toolkit 
that allows for different inference methods including 
those discussed in (Aly and Vrana, 1977; Mamdani, 
1977; Mizumoto, et all, 1979). We examine the 
performance of both FuzzyClips and FuzzyJ Toolkit 
shells and compare their results with that of our shell 
using the four observations detailed in Section 4. For 
this purpose, the English-translation of the data in 
Table 12 is used. 
 
5.1 A comparison with fuzzyCLIPS 

Mamdani's inference methods are the most 
commonly used in commercial fuzzy expert system 
shells. It has been observed that inference methods in 
(Mamdani, 1977) do not satisfy human intuition 
(Fukami et al. 1980; Mizumoto, 1982). We 
demonstrate this observation by examining the 
behavior of FuzzyCLIPS with the rules and fuzzy 
variables of discussed in Section 4. 
 

Table 12. The only criteria satisfied by mamdani's methods 

Criterion I 
Ant1: IF x is A Then y is B 
Ant2: x is A 
Cons: y is B  

Criterion II-2 
Ant1: IF x is A Then y is B 
Ant2: x is very A 
Cons: y is B 

 
The definitions of the fuzzy variables and the fuzzy 
rule using FuzzyCLIPS are shown in Figures 14 to 
18 below: 
 

(deftemplate x ;definition of fuzzy variable 
‘Machine cessation’ 
   15 45 ;Universe of Discourse 

((low (16 0.1) (17 0.3) (18 0.5) (19 1) (20 
0.6)   (21   0.5) ( 22 0.3) (23 0.1)) 
(medium (21 0.1) (22 0.2) (23 0.4) (24 0.5) 
(25 0.6) (26 0.7) ( 27 0.6) (28 0.5) (29 
0.4)(30 0.2)(31 1)) 
(high (31 0.1) (32 0.2) (33 0.3) (34 0.4) (35 
0.5) (36 0.6) ( 37 0.7) (38 0.1) (39 0.7)(40 
0.6)(41 0.5) (42 0.4) (43 0.3) (44 0.2) (45 
0.1))     

  ) 
) 

 
Fig. 14. Variable x definition using fuzzyCLIPS 

 
 

 
(deftemplate y ;definition of fuzzy variable 
‘Absence of professionals’ 
   4 16 ;Universe of Discourse 
  (  
    (low   (5 0.2) (6 1) (7 0.2) ) 
    (medium  (7 0.4) (8 1) (9 1) (10 1) (11 1) 
(12 1)) 
    (high (11 0.2) (12 0.5) (13 1) (14 0.5) (15 
0.2))  
  ) 
) 

 
Fig. 15. Variable y definition using fuzzyCLIPS. 

 
 

(deftemplate z ;definition of fuzzy variable 
‘Weather changes’ 
   10 90 ;Universe of Discourse 
  (  
    (low   (10 0.2) (20 0.6) (30 1) (40 0.2)) 
    (medium   (40 0.7) (50 1) (60 0.7) ) 
    (high (70 0.7) (80 1) ) 
  )) 

 
Fig. 16. Variable z definition using fuzzyCLIPS. 

 
 

(deftemplate cons ;definition of fuzzy 
variable ‘Conclusion’ 
   20 100 ;Universe of Discourse 
  (  

 (low   (25 0.2) (30 0.5) (35 0.7) (40 
1) (45 0.7) (50  0.5) (55 0.2) ) 

    (medium   (55 0.5) (65 1) (70 0.6) (75 0.5) 
) 
    (high (80 0.2) (85 1) (90 0.5) (95 0.2) ) 
  ) 
 
) 

 
Fig. 17. Conclusion definition using fuzzyCLIPS. 

(defrule r1 ; a rule that matches and asserts 
fuzzy facts 
(x low)  (y high) (z high) 
=> 
(assert (cons high) ) 
) 

 
Fig. 18. Fuzzy rule definition using fuzzyCLIPS. 

 



38 Mathkour et al.: The Development of a Bilingual Fuzzy Expert System Shell 
 

 

FuzzyCLIPS was run several times with the same 
observations in Section 4 and the results are as 
follows (Figures 19 to 22): 

 Observation 1 : X is Low , Y is High and Z 
is high. 

 
FuzzyCLIPS gives the expected result according to 
Criteria I in Table 9. This is natural and expected as 
all observations match all antecedents. 

 

 
 

Fig. 19. FuzzyCLIPS result for Observation 1. 
 

 Observation 2: X is not low, Y is not high 
and Z is not high. 

When the antecedents contain the NOT hedge, 
FuzzyCLIPS yields a fuzzy set that cannot be 
mapped to a linguistic expression. This is expected 
as Mamdani's methods do not satisfy Criterion IV-1 
and Criterion IV-2 of Table 9. 

 

 
 

Fig. 20. FuzzyCLIPS result for observation 2. 
 

 Observation 3: X is very low, Y is very 
high and Z is very high. 
 

FuzzyCLIPS gives the expected result according to 
Criteria II-2 in Table 9. 

 

 
 

Fig. 21.  FuzzyCLIPS result for observation 3. 
 

 Observation 4: X is more or less  low, Y is 
more or less  high and Z is more or less 
high. 

The resulting fuzzy set cannot be mapped to a 
linguistic expression. From the result shown in figure 
9 it is clear that Mamdani's methods do not satisfy 
criterion III. 
 

 
 

Fig. 22.  FuzzyCLIPS result for observation 4. 
 
5.2 Comparison with FuzzyJ Toolkit 

FuzzyJ Toolkit is a set of Java classes that 
provide the capability to handle fuzzy concepts and 
reasoning (Orchard,2001). It allows for different 
inference methods including those in (Aly and 
Vrana, 2006; Mamdani, 1977). We examine the 
behavior of FuzzyJ Toolkit using the rules and fuzzy 
variables of Section 4. Figures 23 to 27 show the 
fuzzy variables and fuzzy rule definitions. 
 
 



 J. King Saud University, Vol. 21, Comp. & Info. Sci, Riyadh (2009/1430H.)          39 

 

  

 
 

Fig. 23.  Definition of fuzzy variable x using fuzzyJ toolkit. 
 

 
 

Fig. 24.  Definition of fuzzy variable y using fuzzyJ foolkit. 
 

 
 

Fig. 25. Definition of fuzzy variable z using fuzzyJ toolkit. 



40 Mathkour et al.: The Development of a Bilingual Fuzzy Expert System Shell 
 

 

 
FuzzyJ was run several times with the same 
observations in Section 4 and with the inference 
method set to Larsen's inference method. The results 
of the inference are as follows: 
 

 Observation 1 : X is Low, Y is High and Z is 
High. 

The resulting fuzzy set is {0/70, 0.2/75, 0.5/80, 
1/85, 0.5/90, 0.2/95, 0/100}. Here the result given 
by FuzzyJ is "high" which is natural as all the 
observations match the all the antecedents of the 
fuzzy rule.  

 

 
 Observation 2: X is not low, Y is not high and 

Z is not high. 
The resulting fuzzy set is {0/70 ,0.1/75, 0.25/80 
,0.5/85, 0.25/90 ,0.1/95, 0/100}. This result could 
not be mapped to a linguistic expression although it 
is rather close to the fuzzy set "high". 

  
 Observation 3 & Observation 4:  
The resulting fuzzy set is {0/70, 0.2/75, 0.5/80, 
1/85, 0.5/90 ,0.2/95, 0/100}. Here the result given 
by FuzzyJ is "high". Notice that this is the same 
result when no hedges were used. It is obvious that 
the use of the hedge "very" and the "more or less" 
hedge had no effect on the result.  In our shell the 

 
 

Fig. 26.  Definition of fuzzy variable conclusion using fuzzyJ toolkit. 

 
 

Fig. 27. Definition of the fuzzy rule using fuzzyJ. 



 J. King Saud University, Vol. 21, Comp. & Info. Sci, Riyadh (2009/1430H.)          41 

 

hedges were recognized through the calculation of 
the implication criteria of Table 9. 

 
6. Conclusion 

 
In this paper, we discussed the development of 

our own bilingual fuzzy expert system shell. In the 
process, we have examined, evaluated, and 
compared various fuzzy expert system shells that 
adopt different inference methods for the sake of 
identifying desirable features and examining their 
performance. Our shell was developed using 
NetBeans 4.1 IDE.  It has an Arabic user interface as 
well as an English user interface. The inference 
engine is a backward chaining inference engine. It 
uses the implication methods Rs, Rg, Rss, Rgg, Rgs 
and Rsg.  Several tests have been performed on this 
shell to ascertain its proper functionality. Some of 
the tests have given the expected results that reflect 
human intuitions. Few tests have given results which 
are very close to the expected outcome. We observe 
that when the membership function of fuzzy values 
covers a wide range from 0 to 1, the shell produces 
more accurate results. Experimental results for our 
shell have been reported and analyzed. A comparison 
of the performance of our shell with other shells such 
as FuzzyCLIPS and FuzzyJ has also been discussed. 
We are in the process of extending the shell to allow 
for the processing of fuzzy terms in other natural 
languages.   

References  
 
Aly, S. and Vrana, I.. " Toward efficient modeling of fuzzy expert 

systems: a survey" Agric. Econom. - Czech, (2006), 456-460. 
Bandler, W. and Kohout, L. “Fuzzy power sets and fuzzy 

implication operators”, Fuzzy Sets and Systems, 4, (1980), 
13-30. 

Council, C. and Orchard, R.  “Fuzzy reasoning in jess: The 
fuzzyj toolkit and fuzzyjess”, The Proceedings of 3rd 
International Conference on Enterprise Information Systems 
(ICEIS 2001), Setubal, Portugal, (2001). 

Forgy, C.L. “The OPSS User Manual”, Technical Report, CMU-
CS-81-135, Computer Science Department, Carnegie-Mellon 
University, Pittsburgh, (1981). 

Friedmann-Hill, E.J.. “Jess, The Java Expert System Shell”, 
http://herzberg.ca.sandia.gov/jess/, (1999). 

Fukami, S.; Mizumoto, M. and Tanaka, K. “Some 
Considerations on Fuzzy Conditional Inference”, Fuzzy Sets 
and Systems, 4, (1980), pp.243-273. 

Ganoud, A.; Ali, H. and Ibrahem, S. "Studying the Influence of 
random Factors on the Planning of Building Work", Tishreen 
University Journal for Studies and Scientific research- 
Engineering Science series, 27(3), (2005). 

Giarratano, J.C. “The CLIPS User’s Guide”, 
http://www.twine.com/twine/11lslvj57-19d/clips-expert-
system-shell, (1998). 

Kent, J.R. and Denholm, P. “ESB-96 – A Graphical KBS 
Development Tool”, in Europal'90, Proceedings of the First 
European Conference on the Practical Application of Lisp. 
Dorking, March (1990). 

Ishihara, S.; Ishihara K.; Nagamachi, M. and Matsubara, Y. 
“An Automatic Builder for a Kansei Engineering Expert 
System using self- Organizing Neural Networks”, Int. Journal 
of Industrial Ergomonics, 15(1), (1995), 13-24. 

Kelmet, E. and Slany, W. “Fuzzy Logic in Artificial 
Intelligence”, Proceedings of the 8th Austrian Artificial 
Intelligence Conference, FLAI '93, Linz, Austria, (1993). 

Leung, K. S.; and Lam, W. "Fuzzy Concepts in Expert 
Systems",IEEE Computer, 21(9), (1988), 43-56. 

López-Ortega, O. " Java Fuzzy Kit (JFK): A shell to build fuzzy 
inference systems  according to the generalized principle of 
extension", Expert Systems with Applications, (2006). 

Mamdani, E.H. “Application of fuzzy logic to approximate 
reasoning using linguistic systems”, IEEE Trans. Comput. 26, 
(1977), 1182-1191. 

Mathkour, H.; Al-Turaiki, I. and Touir, A. “The Development 
of an Arabized Fuzzy Expert System Shell” in int. conf on 
Information Management and Engineering ICIME, (2009). 

 
Mathworks, Inc. Fuzzy Logic Toolbox Use's Guide version 2, 

(1999). 
Mizumoto, M.; Fukami, S. and Tanaka, K. “Fuzzy conditional 

inference and fuzzy inference with fuzzy quantifiers”. in: 
Proc. of 6th Int. Conf. on Artificial Intelligence, Tokyo, 
(1979), 589-591. 

Mizumoto, M.; Fukami, S. and Tanaka, K. “Some methods  of 
fuzzy reasoning”, in: M.M. Gupta et al., ads., Advances in 
Fuzzy Set Theory and application (North-Holland, 
Amsterdam, (1979), 117-136. 

Mizumoto, M. Note on the arithmetic rule by Zadeh for fuzzy 
conditional inference, Cybernetics and Systems, 12, (1981), 
247-306. 

Mizumoto, M. “Fuzzy inferences under max-  composition in the 
compositional rule of inference”, in: M.M. Gupta et al., ads., 
Fuzzy Information and Decision Processes, North-Holland, 
Information Sciences, 27, (1982), 183-209. 

Mizumoto, M.; Fukami S. and Tanaka, K. “Several methods for 
fuzzy conditional inference”, in: Proc. of IEEE Conf. on 
Decision & Control, Florida, (1979), 777-782. 

Orchard, R.  “NRC FuzzyJ Toolkit for the Java™, Platform 
User’s Guide” [online]. National Research Council of 
Canada. Available 
from:http://ai.iit.nrc.ca/IR_public/fuzzy/fuzzyJDocs/index.ht
ml, 2001. 

Orchard, R. "Fuzzy Reasoning in Jess: The FuzzyJ Toolkit and 
FuzzyJess", Proceedings of the ICEIS 2001, Third 
International Conference on Enterprise Information Systems, 
Setubal, Portugal.  (2001), 533-542.  

Orchard, R. FuzzyClips version 6.10d User's Guide,National 
Research Council of Canada, (2004). 

Pan, J. FuzzyShell User's Manual,(1996). 
Philip, G., "A case study in selecting an expert system shell", 

Journal of Systems Management, (1991). 
Shalfield, R. FLINT Reference, Logic Programming Associates 

Ltd.,(2005). 
Siler, W. and Buckley, J. Fuzzy Expert Systems and Fuzzy 

Reasoning, John Wiley & Sons, Inc., New Jersey, (2005). 
Vasey P.; Westwood D. and Johns N. “Flex Expert System 

Toolkit”, Logic Programmers Associates Ltd. (1996). 



42 Mathkour et al.: The Development of a Bilingual Fuzzy Expert System Shell 
 

 

Willmott, R., “Two fuzzier implication operators” in the theory of 
fuzzy power sets, Fuzzy Sets and Systems, 4, (1980), 31-37. 

Zadeh, L.A. Calculus of fuzzy restriction, in: L.A. Zadeh et al., 
Eds. Fuzzy Sets and Their Applications to Cognitive and 
Decision Processes (Academic Press, New York, (1975) 1-39. 

Zadeh, L.A. "From Computing with Numbers to Computing with 
Words− From Manipulation of Measurements to 
Manipulation of Perceptions", iee transactions on circuits and 
systems, Fundamental theory and applications, 45(1), 
JANUARY (1999), 105-119. 

Zadeh L.A. “Fuzzy sets”   information and control 8, (1965), 338-
353.  

Zadeh, L. A. "Why the Success of Fuzzy Logic is not Paradoxial", 
IEEE Expert, 9, (1994), 43-45. 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 

 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 



 J. King Saud University, Vol. 21, Comp. & Info. Sci, Riyadh (2009/1430H.)          43 

 

 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 

 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 

 تطوير صدفية ثنائية اللغة لبناء النظم الخبيرة الضبابية
 
 

 عامر عبداهللا طويرو  حسن إسماعيل مذكور، إسراء محمد الطريقي
 

 قسم علوم احلاسب، كلية علوم احلاسب و املعلومات، 
 العربية السعوديةجامعة امللك سعود، الرياض،اململكة 

mathkour@ksu.edu.sa 
 

 )م٢/٣/٢٠٠٩؛ وقبل للنشر يف م٢١/١٢/٢٠٠٨(قدم للنشر يف 
 

يستخدم املنطق الضبايب يف بعض النظم اخلبرية حلل مشكالت من احلياة العملية واليت تتسم بالغموض. إذ . ملخص البحث
استخدام املنطق الضبايب لربجمة التطبيقات اليت تعتمد على احلدس البشري من خالل بناء صدفيات لتطوير النظم من املمكن 

اخلبرية الضبابية من أحل التعامل مع املشكالت الضبابية و الغري واضحة. و توفر هذه الصدفيات تشكيلة من الدوال ميكن 
االت و التطبيقات مثل التطبيقات الطبية، استخدامها لتسهيل تطوير نظم خبرية ضبابية للتعا مل مع مشكالت يف خمتلف ا

واهلندسية، واملالية. ال تتوفر حسب علمنا صدفيات برجمة ضبابية، لبناء النظم اخلبرية الضبابية، مطورة يف األصل للتعامل مع 
لضبابية ثنائية اللغة. اهلدف من هذه الصدفية اللغة العربية. تصف ورقة العمل هذه تطوير وجتربة صدفية لبناء النظم اخلبرية ا

الربجمية هو استخدامها كأداة حبثية ملطوري النظم اخلبرية الضبابية يف البيئات ثنائية اللغة كما هو احلال يف العامل العريب حيث 
مع املصطلحات تكون استخدامات املطورين متعددة اللغات بسبب نظام تعليمهم وبيئة عملهم. تسمح الصدفية بالتعامل 

الضبابية يف اللغتني العربية واإلجنليزية. وتعترب صدفية برجمة عامة األهداف توفر للمستخدمني القدرة على تطوير نظم خبرية 
ضبابية عربية وإجنليزية باستخدام واجهة مستخدم مبسطة. وتقوم بتطبيق طرق االقتضاء يف حماكاة للحدس البشري. مت عند 

ارنة عدة صدفيات لتطويرو برجمة النظم اخلبرية الضبابية لتحديد نقاط القوة والضعف للربجميات املتوفرة.  التطوير دراسة و مق
كما مت إعداد تقرير عن جتربة و اختبار الصدفية املطورة ومقارنة أدائها مع الصدفيات املتوفرة واليت تستخدم طرقًا اقتضائية 

 خمتلفة.



44 Mathkour et al.: The Development of a Bilingual Fuzzy Expert System Shell