Content uploaded by Emad Khorshid

Author content

All content in this area was uploaded by Emad Khorshid on Jun 28, 2018

Content may be subject to copyright.

Int. J. Design Engineering, Vol. X, No. Y, xxxx 1

Copyright © 200x Inderscience Enterprises Ltd.

A new optimal Arabic keyboard layout using genetic

algorithm

E. Khorshid*, A. Alfadli and M. Majeed

Mechanical Engineering Department,

Kuwait University,

P.O. Box 5969 Safat, Kuwait

E-mail: Khorshid@kuc01.kuniv.edu.kw

E-mail: alfadlia1@yahoo.com

E-mail: mmajeed@kuc01.kuniv.edu.kw

*Corresponding author

Abstract: A new design of Arabic keyboard layout for convenient typing and

effective use of keyboard is proposed. The design methodology is based on

ergonomic criteria that evaluate and compares different keyboard layouts for

optimal characters distribution. The proposed design criteria are simulated by a

mathematical model that uses:

1 best distribution of the typing effort among the ten fingers of the typist

2 accessibility of commonly used characters

3 various other factors as constraint equations to reach an optimal keyboard

layout that is ergonomically improved.

This task is carried out using a genetic algorithm based optimisation

framework. The new keyboard layout shows better ergonomic performance

than the present keyboard being used in Arabic world with a 36.3%

improvement.

Keywords: keyboard design; genetic algorithm; ergonomic design; global

optimisation.

Reference to this paper should be made as follows: Khorshid, E., Alfadli, A.

and Majeed, M. (xxxx) ‘A new optimal Arabic keyboard layout using genetic

algorithm’, Int. J. Design Engineering, Vol. X, No. Y, pp.000–000.

Biographical notes: Emad Khorshid received his PhD in Mechanical

Engineering at the University of Wisconsin-Madison. Since 1998, he has been

a member of the Mechanical Engineering Department at Kuwait University.

His main research interests are design and evaluation of traffic calming devices,

optimal design of mechanical systems, complex system analysis and whole

body vibration injuries. He has published more than 23 articles in archival

journals and conference proceedings.

AbdulAziz Alfadli received his Master in Mechanical Engineering at Kuwait

University. Since 2001, he is working as Teaching Assistant at the Mechanical

Engineering Department at Kuwait University. His main research interest are

rotor dynamics, systems design optimisation, and vibration.

Majed Majeed received his PhD in Mechanical Engineering at the Virginia

Polytechnic Institute and State University. Since 2005, he has been a member

of the Mechanical Engineering Department at Kuwait University. His main

research interests are structural mechanics, mechanics of composite materials,

vibrations and control. He has published more than 12 articles in archival

journals and conference proceedings.

2 E. Khorshid et al.

1 Introduction

The distribution of the various characters on a keyboard has profound impact on the rate

of typing and typing performance. If frequently used characters are not easily accessible,

the rate of typing will dramatically drop down. An ill-designed keyboard

disproportionately places high load on the weaker fingers leading to typing fatigue, which

possibly causes musculoskeletal injuries. To avoid such injuries and enhance typing rate,

the optimal characters arrangement are sought. For English, European, and Hindi

keyboards, a considerable amount of research was invested towards finding the best

possible arrangement of characters (Pollatschek et al., 1975; Anil et al., 2000; Wagner et

al., 2003). On the other hand, to the knowledge of the authors, there is no firm proof of

whether currently used standard Arabic keyboard is truly optimal and what optimisation

methods have been used. For this reason and the fact that there are other possible

ergonomically optimised layouts has motivated the authors to further investigate an

alternative optimal Arabic keyboard layout based on ergonomic standards.

The optimisation of English keyboard layout based on the typing speed and rapid

learning of the typing system were conducted by Norman and Fisher (1982). Pollatschek

et al. (1975), and Burkard and Offermann (1977) investigated other English layouts using

a broader criterion but did not rely on a realistic model that would have addressed the

problem completely. In general, many optimisation methods were used in keyboard

layout optimisation including, but not limited to, genetic algorithm (GA) (Anil et al.,

2000), simulated annealing (Light and Anderson, 1993), and ant colony optimisation

(Wagner et al., 2003).

Besides the work of Idlebi and Mrayati (1990), who attempted to design a more

efficient Arabic keyboard based on statistical approach no other scientific work was

found and been devoted to obtain an optimal keyboard arrangement for Arabic language

that takes into consideration ergonomic criteria. It is believed that current Arabic

language keyboard layout has been designed by simply adopting English-based computer

systems for Arabic language users. This, to some extent, is not satisfactory because, by

nature, the Arabic language is much different than English. For example, Arabic is

written from right to left and the frequency of letter appearance is different.

The main objective of the proposed study is to optimise the keyboard layout using

GA combined with performance criteria based on six ergonomic factors. The six criteria

used to optimise the Arabic keyboard, given by Yannou and Hossenlopp (2000), are

detailed later in the paper. For comparison purposes, the proposed Arabic keyboard is

compared to the current standard Arabic Keyboard.

2 Optimisation problem formulation

The process of typing any given text can be decomposed into a sequence of key strikes,

namely, a sequence of single or consequent characters usually known as monographs or

digraphs. A monograph is an isolated key strike carried out in the process of typing a text

while a digraph is a sequence of two consecutive keys strikes. Most of previously

published work on keyboard design used criteria based on the distinction between

monograph and digraphs as discussed by Anil et al. (2000) and Wagner et al. (2003).

Based on this distinction, it may be therefore useful to establish statistical data for the

appearance frequency of monographs and digraphs in Arabic literature that would serve

A new optimal Arabic keyboard layout using genetic algorithm 3

as a good ground for optimisation. For this purpose, let fmi and fdi present the statistical

appearance frequencies of monograph mi and digraph di, respectively. The source of these

monograph and diagraph data can be obtained from the available electronic literature

with a wide variety of topics and subjects. The literature sources could vary from

newspapers, electronic books, scientific journals, etc.

Wagner et al. (2001), used Le Monde, Der Spiegel, and USA Today newspapers for

French, German, and English text sources, respectively. On the other hand, Deshwal and

Deb (2004) used various text sources from the internet (15 MB of text ISCII format) for

an Indian keyboard design. Mrayati (1990) showed marginal difference in both

monographs and digraphs frequencies for texts resources taken from three different Arab

countries. Some statistical studies did use the holy Quran as a source to calculate the

frequency of the monographs and the digraphs (Mousa, 1983), which is adopted in the

current work. The reason behind this selection is that, as believed by all Muslims, the

holy Quran is the miracle that challenged the Arabs in their language; nevertheless,

electronic versions sassily accessible. Muslim scholars have considered the linguistic

miracle of the holy Quran as perhaps the most important miraculous aspect of the holy

Quran and studies by Draz, (2001) and Mahmoud, (2004) show that indeed the holy

Quran is a true linguist challenge and it is definitely the one that had the most influence at

the time of the Prophet Muhammad, may the mercy and blessings of God be upon him

(Zarabozo, 2007).

Figure 1 Monograph frequency for Arabic letters in the Quran (see online version for colours)

A computer program was developed for counting the monograph and diagraph from a

digital Quran script and the results are shown in Figure 1 and Figure 2. These figures

demonstrate the statistical frequent appearance of the monograph and digraph

respectively for the main 34 Arabic letters. The y-axis in both figures represents the

monograph and the diagraph scaled to the total number of letters. As can be seen from

4 E. Khorshid et al.

both figures, and as expected, letter ‘ا’ has the highest frequency among all letters. This

result reflects the fact that letter ‘ا’ is the most used letter in the Arabic language.

Furthermore, it is observed from Figure 2 that the digraph ‘ﻞا’ has the highest digraph

frequency among all digraphs. Again, this result is expected because this particular

digraph is the definition article in the Arabic language likewise the article ‘The’ in the

English language.

Figure 2 Diagraph frequency for Arabic letters in the Quran (see online version for colours)

2.1 Representation of the keyboard problem

A keyboard is designed to translate handwritten/typewritten strings of letters into

electronic format using a set of character keys distributed over a board. Analogous to

typewriter, the keys are distributed in a way that can be described by rows and columns

with variety of rules and arrangements. Figure 3 shows the layout of the standard Arabic

keyboard in use. For the problem to be addressed properly, the keyboard is geometrically

represented by rows and columns of ergonomically spaced character-keys and each

character is described by four indices, namely (x1, x2, x3, x4), with the following rules

applied:

• A left hand is represented by one and a right hand is represented by zero.

• Column indices vary from zero to seven for the left hand and from zero to eight for

the right hand.

• Row indices vary from zero to five, zero represents the top row, which will not

contain any letter while five represents the bottom row, which contains keys for

(Space, Alt, and Ctrl). On the other hand, row 3 represents the reference row of

which the typist rests at.

A new optimal Arabic keyboard layout using genetic algorithm 5

Figure 3 Standard Arabic keyboard (see online version for colours)

The indices x1, x2, x3, and x4 are the shift key, side, column number, and row number.

Using the above mentioned rules, any character of the keyboard in Figure 3 can be

mapped to a unique position using the designation (x1, x2, x3, x4). For example and as can

be seen in Figure 4, the letter ‘ﺖ’ is mapped as (0, 0, 2, 3). Here the first index, zero,

denotes the use of the shift key which is turned off in this case. The second index

represents the right side location (right hand is used for the letter ‘ﺖ’). The third and the

forth indices, two and three, in this case, represent the second column and the third row,

respectively. Note that the number 18 shown on the top of letter ‘ﺖ’ in Figure 4 indicates

the mapped relation between the location 18 and the (0, 0, 2, 3).

Figure 4 Standard Arabic keyboard with indices of rows, columns and the finger (see online

version for colours)

Figure 4 shows the mapped position of the standard Arabic keyboard, where the Arabic

keys that contain the Arabic letters/characters (from ﺪ to ئ) are given numbers from (one

to 34); starting with number 1 for character (ﺪ) and ending with number 34 for character

(ئ). Note that the key of letter (ذ) is not located in series with the other letters therefore it

is numbered as it belongs to the second row in the keyboard. During the optimisation,

each letter/character is initially given an index number equal to the index location of the

key. The location index vector is fixed while the Arabic characters index vector changes

according to the suggested position. Each character (element) in the index vector

representing the Arabic characters is mapped to 4D array (x1, x2, x3, x4) that changes

during the optimisation process. During each iteration step, the index vector representing

the Arabic characters is compared to the index vector representing the

remove and

add ,

6 E. Khorshid et al.

key-location and each letter is relocated to a new position producing a new keyboard,

which is then evaluated by the ergonomic criteria. For example, as will be shown later,

the initial location of the letter (ﺪ) before optimisation is equal to one and the final

location after applying the ergonomic constraints is changed to 28.

Since the objective here is to rearrange the Arabic letters in the standard keyboard in

an optimal layout for better typing performance and efficiency, the skeleton of the

standard Arabic keyboard in terms of the number of rows, number of columns, left hand,

and right hand fingers will not change. Once the standard Arabic keyboard is translated to

the abstraction (array) keyboard of Figure 4, the objective function can then be applied

easily and each letter can be mapped to a specific key location on the board. To keep

track of letter positions, the keys are numbered starting from the right side of the

keyboard sweeping to the left.

2.2 Objective function

In the current study, the objective function, known also as the cost function, represents

the quality of the keyboard layout in the sense that it should:

1 permit for minimum typing effort

2 maximise typing speed

3 reduced typing errors

4 ease of learning and memorising character locations.

In this work, it is proposed that each keyboard arrangement is evaluated based on the six

criteria listed by Yannou and Hossenlopp (2000), namely, load location, number of

keystrokes, hand alternation, consecutive usage of the same finger, avoid large steps, and

hit direction. The final score of the keyboard is determined as a weighted linear

combination of theses six individual criteria. The following sections explain each one of

these six criteria and their mathematical representation.

2.2.1 Load distribution

It is obvious that while typing, the total load on the fingers is constant and since each

finger of the hand has definite strength, some keys are less accessible than others. It

would be highly desirable if this total load can be distributed among the fingers in

proportion of their relative strength.

Mathematically, an ideal load distribution can be assigned between all the

monographs while the performance of any keyboard can be evaluated by determining the

deviation between the actual load distribution for this keyboard and the ideal load

distribution (Eggers et al., 2001). Initially, a load weight factor (wL) can be assigned to

each hand representing a possible difference in their performance and endurance. To

divide the total load equally between the left and right hands this value is chosen to be

50% for each hand. Therefore, each column receives a ratio which represents the relative

agility and endurance of the fingers. Also, each row receives a ratio representing the

relative accessibility. These ratios are presented on Table 1 which is derived from the

work by Eggers et al. (2001). The data in Table 1 are related to ergonomic load on the

typist fingers regarding of the letter (Arabic or English) assigned to the specified key.

criteria

remove

on the board

A new optimal Arabic keyboard layout using genetic algorithm 7

Note that row number 0 in Table 1 is not used for the current layout design since it

represents the function keys (F1 to F12) of the standard keyboard (see Figure 3). The

same assumption is used for the raw number 1 since it represents the numeric numbers

which will not be altered for the new proposed layout design. Note that in Figure 5 the

maximum ideal load distribution max 0.050176

()

=

opt

m

f is given for a monograph that had been

positioned in the third row and the fourth column. Also, since the ideal load distributions

are too small, they were normalised to a maximum of 600.

Table 1 Ideal load distribution for the typist fingers (Eggers et al. 2001)

No. Row (%) Column (%) Finger type

0 10.87 15.38 Little

1 13.04 10.26 Little

2 15.22 15.38 Little

3 43.48 23.08 Little

4 10.87 17.95 Ring

5 6.52 6.41 Middle

6 - 5.13 Index

7 - 3.85 Index

8 - 2.56 Thumb

Figure 5 Ideal load distribution for the left hand (see online version for colours)

Once these ratios of Table 1 are obtained, the optimal load distribution i

opt

m

f

for a

monograph, m = (hand, row, column), can be calculated by multiplying the respective

ratios in Table 1 for a given row and column and then multiplying the result by the ratio

given for the hand being used for typing this monograph. For example, to calculate the

optimal load distribution for the letter ‘أ’ in the standard Arabic keyboard which has a

position of (1, 1, 3) in the indices format. Hence, the first index indicates that the type of

8 E. Khorshid et al.

the hand used to print this letter (zero is given for the right hand while one is given to the

left hand). The remaining two indices indicate that the letter ‘أ’ is positioned in the first

column (little finger) and the third row. Therefore, from Table 1, the optimal load

distribution i

opt

m

f

is calculated as

(weighted row) * (weighted column)

i

opt

L

m

fw=

أ0.5 *(15.38 /100) * (43.48 / 100) 0.033436.

opt

f==

This criterion used to calculate the variance of the load distribution on a keyboard from

the ideal load distribution given by Table 1. Therefore, the score of load distribution is

given by the following equation

(

)

1

2

1ii

m

i

opt

mm

m

vff

∈Ξ

=−

∑ (1)

where 1

m

m

m

f∈Ξ is the set of all monographs represented in Figure 1.

2.2.2 Key number

In order to have an efficient keyboard, the number of keystrokes needed to produce a

given text has to be minimised. Therefore, the score v2 of this criterion is presented by the

ratio of the total number of characters in a given text to the total number of keystrokes

necessary to produce the text. Generally, for different solutions of the KAP (KAP refers

to keyboard arrangement problem as defined in many literatures in the field), this score

has no difference. Hence, the numbers of keys necessary for each character are fixed. In

another word, since the shift key is not used, the number of keys to perform typing is

fixed.

2.2.3 Hand alternation

Typing process will be efficiently fast and more comfortable provided that subsequent

keys are hit by opposite hands. Having this granted will ensure that while one hand is in

the process of typing the first key the other hand has moved to the next key position.

Therefore, the performance of this score is calculated by summing up the frequency of

digraphs which are typed using one hand only. Since this score represents the summation

of all digraph that are typed using the same hand and this summation must be minimised,

In another word any digraph is produced as a combination of two consecutive characters

when these two characters are located in a keyboard such that they are hit by using

fingers of the same hand. This will reduce the typing process. For example the digraph

‘ﻞأ’ in the keyboard given by Figure 3 is produced by using fingers from both hands and

so that this digraph does not belong to this category. On the other hand the digraph ‘ﻢأ’ is

produced by using fingers of one hand (the right hand). Mathematically, it is easy to

capture this set by using the mapping method (hand, row, column) so when the first index

for the two characters is the same this indicates that they typed by the same hand. This

rule is called the hand alternation rule (HAR) which states that consecutive keys must not

hit by the same hand. Mathematically, this performance is formulated as follows:

A new optimal Arabic keyboard layout using genetic algorithm 9

3

3i

d

i

d

d

vf

∈Ξ

=∑ (2)

where 3

d

i

dΞ is the set of all digraphs that are typed using one hand only and i

d

f

is the

frequency of the HAR of each digraphs.

2.2.4 Consecutive usage of the same finger

Typing process might be slowed down if two consecutive keys are hit by the same finger.

Therefore, in this criterion the above HAR is also applied for the fingers. The score of

this criterion is calculated by summing up the frequencies of all digraphs that are typed

using the same finger and multiplying each of them with distance function. The greater

distance between the two keys of a digraph, the more inconvenience for a consecutive

usage. Mathematically, this index is formulated as:

4

4()

i

d

i

di

d

vfdistd

∈Ξ

=∑ (3)

where 4

d

Ξ is the set of all digraphs that are typed by using the same finger of one hand

and dist (di) is the Manhattan distance function which is given by:

21 12

()| || |

i

dist d c c r r

=

−+− (4)

where c1 and c2 are the respective columns of the two consecutive keys needed to produce

the diagraph and r1 and r2 are the corresponding rows.

For example, suppose that the diagraph ‘ﻢآ’ is hit by the same finger, then

dist (d) = abs(5–6) + (3–3) = 1 and fdi = 59.74250e-6 (from diagraph data), thus,

v4 = 59.74250e-6.

2.2.5 Avoid big steps

When two consecutive keys hits by the same hand, great distances which need

uncomfortable hand postures should be avoided. This exclusively occurs for keys whose

vertical distance is grater than or equal to one. Therefore, relevant set 5

d

Ξ

is the set of

digraphs which are typed using the same hand, but not the same finger and the vertical

distance between the two keys is greater than or equal to one row. Mathematically, this

index is formulated as following

5

5(, ) i

d

i

d

d

vkijf

∈Ξ

=∑ (5)

where k(i, j) is the weight coefficient of the i and j, first and second finger, respectively.

The values of these coefficients were taken from Wagner et al. (2001) and are presented

in Table 2.

It is clear from Table 2 that the coefficient increases with the following pairs of

fingers (mid-index) has a value of five, (little-index) has a value of six, (little-mid) has a

value of seven, (index-ring) has a value of eight, (mid-ring) has a value of nine, and

(ring-little) has a value of ten.

10 E. Khorshid et al.

Table 2 The values of weight coefficients used for calculating v5

Fin

g

e

r

Thumb Index

M

id

R

in

g

L

ittle

Thumb 0 0 0 0 0

Index 0 0 5 8 6

Mid 0 5 0 9 7

Ring 0 8 9 0 10

Little 0 6 7 10 0

2.2.6 Hit direction

Ergonomic researches have shown that the preferred hit direction for digraphs that are

typed using one hand only is from the little finger towards the thumb (Anil et al., 2000

and Eggers et al., 2001). Since this is the natural finger movement for most people. The

relevant set 6

d

Ξ is therefore the set of digraphs which are typed using one hand and

whose hit direction is not the preferred one. For example, the diagraph ‘نﻤ’ in the

standard keyboard (Figure 4) is typed by one hand and used two different fingers firstly

the ring finger and secondly the mid finger. Therefore, the natural finger movement of

this diagraph does not belong to this category. On the other hand the diagraph ‘حﻤ’ is

typed by one hand and used two different fingers firstly the ring finger and secondly the

little finger and this direction of typing is not the natural movement of fingers, so that this

diagraph is belong to this category. This index can be represented mathematically as

follows:

6

6i

d

i

d

d

vf

∈Ξ

=∑ (6)

2.2.7 Overall score

Once the above partial scores vj (1 ≤ j ≤ 6) j are obtained, the overall score, which is used

to compare different keyboard arrangements, can be calculated by obtaining a weighted

sum of these scores. Since these scores have different relative importance, limits, and

units, they are first turned dimensionless by dividing each one of these scores by the

respective score of a reference keyboard vj,ref. (the current keyboard) and then multiplying

by a relative weight coefficient γj and summed. The values of the weight coefficients γj

are given in Table 3. They were obtained with the help of two specialised agronomists,

using a pair wise comparison method (Refer to Limayen and Yannou, 2002).

Table 3 Relative weight coefficients γj

Index Relative weight (γj) Score of vjref (standard)

Load distribution (v1) 0.45 0.0341

Key number (v2) 0.5 1.0000

Hand alternation (v3) 1 0.4980

Consecutive usage of the same finger (v4) 0.8 0.1401

Avoid big steps (v5) 0.7 1.4516

Hit direction (v6) 0.6 0.1632

Overall score (vref) - 4.0500

v5:

v is italic font

5 is suberscript

A new optimal Arabic keyboard layout using genetic algorithm 11

Finally, the overall score is formulated as follows:

6

,.

1

j

j

jref

j

v

Vv

γ

=

=∑ (7)

Since the objective function of the keyboard arrangement which is given in equation (7)

needs first to obtain the values of the partial scores of the reference keyboard, and since

the reference is assumed to be the standard Arabic keyboard shown in Figure 3, as

mentioned above, the above six ergonomic criteria can be used to calculate the partial

scores and the objective function to be used in the optimisation process.

The optimisation problem can be presented in a standard optimal design problem

form as follows:

: Arabic characters of 34 letters with their monograph and

diagraph frequencies, the standardArabic keyboard.

: ; 1: 34 where represents the index position of the

letter in the keyboard.

ii

Given

Find x i x

i

T

=

: ( ) where is given in equation (7)

: is integer and 1 34

ii

o minimise f x V V

Subject to x x

=

≤≤

(8)

3 Optimisation method

The optimisation problem statement in equation (8) falls under the integer nonlinear

programming, which can be solved using GA method. The GA method is selected as a

solution method because of its robustness and ability to capture the global optimal

solution for integer variables. The GA toolbox in MATLAB is used as the solver engine

(Math Works, 2006), which can handle only continuous variables. Therefore, the first

step in the solution is to modify the GA optimisation code to deal with integer variables.

The modification was done on the mutation, creation, and population functions. The

original codes for the Creation function are used to generate individual populations. The

original codes for the mutation function are used to generate children. In both creation

and mutation steps the function ‘rand’ is used to generate random continuous

individuals’ population and children, respectively. Thus, to generate random integer

individual’s population and children the rand function is changed to randint function.

With this modification the handling of integer variable is possible with the GA toolbox.

All computations were conducted on Intel® Pentium®4 machine with 1 gigabyte memory

of 333 MHz speed.

4 Discussion and analysis of the results

In this section, several computational simulations are conducted in an attempt to find the

best optimal solution (global optimal). The effect of several parameters in the GA method

is studied towards finding global optimal solution. Also, the new layouts of the Arabic

language keyboard are compared to the currently available standard Arabic keyboard.

space between standard

and Arabic

Small c letter

12 E. Khorshid et al.

The comparison is based on the six proposed ergonomic criteria of equations (1) through

(6).

Figure 6 Evaluation of the optimal solution for 200 generation and 500 populations (see online

version for colours)

The parameters used for the GA toolbox for Case-1 set as follows: the population size is

500, the number of generations 200, the mutation factor is 0.2, and the crossover factor is

0.7. A sample run shows the relation between the objective function and the number of

iteration is presented in Figure 6. Note that in the top portion of the figure, the y-axis

represents the Fitness value, or cost function of equation (7), of the best solution (and the

average fitness values of all solutions in the population) during each iteration. The lower

plot in Figure 5 represent the optimal solution in each iteration during the solution

process until the final best solution which shown in the figure. As a common trend in the

GA method, Figure 6 shows that the solution starts to converge very fast with the initial

generations and then it gets slower until reaching the final optimal solution of 2.6323.

The CPU time for this optimal solution is 5.32 hours. Figure 7 illustrates the effect of

changing the number of population from 500 to 2,000 and the number of generation is

changed from 200 to 220. The final optimal solution is 2.5162 is reached in 20.2 hours of

the CPU time. This run is defined as Case-2. The final results for the best known optimal

solution for Case-2 are shown in Figure 8.

A new optimal Arabic keyboard layout using genetic algorithm 13

Figure 7 Evaluation of the optimal solution for 220 generations and 2,000 population

(see online version for colours)

Figure 8 The optimised arrangement for Arabic language keyboard (see online version for

colours)

Table 4 shows a comparison between the old keyboard and the newly optimised one. It is

clear that each subjective criterion is improved for the new design. The only exception is

that for the partial score (v2), which is related to the key number, because the shift keys

used with Tashkeel is not used in the present work. This can be added in a future work.

Finally, the greatest improvement in the six criteria is the consecutive usage of the same

finger (v4) as shown in Table 4.

14 E. Khorshid et al.

Table 4 Comparison between the optimal arrangement an the keyboards (equal weights for

both hands)

Index Old arrangement

New arrangement

(weighted) % improvement

Load distribution (v1) 1 0.7977 20.23%

Key number (v2) 1 1.0000 0%

Hand alternation (v3) 1 0.7072 29.27%

Consecutive usage of the

same finger (v4)

1 0.3812 61.88%

Avoid big steps (v5) 1 0.5722 42.78%

Hit direction (v6) 1 0.4988 50.12%

Overall score (vref) 1 0.6370 36.30%

5 Effect of changing the total hand load on the optimal solution

In the previous discussion, the total load was divided equally between the two hands. In

this section, the effect of changing the load distribution on the optimisation process and

the resulting layout is investigated. Suppose that a higher load is given for one hand and

consequently, a lower load for the other hand (e.g., a load factor of 0.7 is given for right

hand while 0.3 is given to the left hand). The results of the new layout are shown in

Figure 9 and Table 5, respectively. It was found that the difference between the current

results and the results from the previous section was only in the load distribution score,

which is obvious. As a result, the load distribution is increased from 0.8 for equally

distributed load to 0.85 for unequally distributed load. This increase has resulted in

overloading the right hand. This shows that the new Arabic keyboard layout is insensitive

to the hand used. Therefore, left-handed users or right-handed users will both experience

the same efficiency with the proposed new layout.

Figure 9 The optimised arrangement for Arabic language for right hand 0.7 and 0.3 for the left

hand (see online version for colours)

A new optimal Arabic keyboard layout using genetic algorithm 15

Table 5 Comparison between standard and optimal keyboards with 0.7 for right hand and 0.3

for left hand load

Index Old arrangement

New arrangement

(weighted) % improvement

Load distribution (v1) 1 0.8491 15.0937

Key number (v2) 1 1.0000 0

Hand alternation (v3) 1 0.7073 29.2747

Consecutive usage of the

same finger (v4)

1 0.3810 61.9048

Avoid big steps (v5) 1 0.5722 42.7821

Hit direction (v6) 1 0.4990 50.1048

Overall Score (vref) 1 0.6395 36.05

6 Conclusions

The possibility of optimising the current standard Arabic keyboard based on ergonomic

criteria for best typing performance and for more convenient and effective typing was

discussed; and a new keyboard layout was proposed. The design methodology was based

on six ergonomic criteria using a GA technique. The use of these six ergonomic criteria

permits to evaluate and compare different keyboard layouts for optimal characters

distribution. The mathematical model used for optimisation was based on:

1 best distribution of the typing effort among the ten fingers

2 accessibility of commonly used keys

3 various other factors as constraint equations to reach an optimal keyboard layout that

is ergonomically improved.

The new keyboard layout shows better performance than the present keyboard being

currently used worldwide. The total improvement based on the ergonomic criteria is

36.3% compared to the current keyboard. It was observed that changing the hand load

distribution has a minor effect on the optimised layout design. The overall effect of

changing from equal to unequal hand load distribution was less than 3% from the overall

score. It is suggested to test the new layout on newly trained typists.

References

Alden, D., Daniels, R. and Kanarick, A. (1972) ‘Keyboard design and operation: a review of the

major issues’, Human Factors, Vol. 14, pp.275–293.

Anil, B., Ramakrishnan, C., Ashish, D., Crouch, D. and Hasan, S. (2000) ‘Optimizing ambiguous

keyboard layout for the English language via genetic algorithm’, Final project in

Computational Intelligence CSCI 8940, Spring semester 2000, Intelligence Center, University

of Georgia, Athens.

Burkard, R. and Offermann, J. (977) ‘Entwurf von Schreibmaschinentastaturen mittels

quadratischer Zuordnungsprobleme’, Zeitschrift f¨ur Operations Research, Vol. 21,

pp.B121–B132.

16 E. Khorshid et al.

Deshwal, S. and Deb, K. (2004) ‘Design of an optimal Hindi keyboard for convenient and efficient

use’, KanGAL Report Number 2003004, IIT, India.

Draz, M. (2001) The Quran: An Eternal Challenge (Markfield, United Kingdom: The Islamic

Foundation, p.69.

Eggers, J., Feillet, D., Kehl, S., Wagner, M. and Yannou, B. (2001) ‘An ant colony optimization

algorithm for the optimization of the keyboard arrangement problem’, Technical Report

CER 01-03 A, August, Laboratoire Productique Logistique, Ecole Centrale Paris.

Idlebi, N. and Mrayati, M. (1990) ‘Design Arabic keyboard layout based on statistical properties of

Arabic characters’, in Computers and the Arabic Language, pp.97–101, MacKay, P.A. (Ed.):

Proceedings of the Arab School of Science and Technology, Hemisphere Publishing Corp.,

New York.

Light, L. and Anderson, P. (1993) ‘Typewriter keyboards via simulated annealing’, AI Expert,

September issue.

Mahmoud, M. (2004) Understanding the Quran: A Contemporary Approach, Amana Publications.

Marsan, C. (1976) ‘Perfectionnements aux claviers de machines à écrire et similaires’, brevet

d’invention no. 76-23323, l’Institut National de la Propriété Industrielle, 30 July, France.

Math works (2006) GA Toolbox.

Mousa, A. (1983) ‘Statistical studies of Arabic language routs and Arabic words’, Proceedings of

the Arab School in Science and Technology, Rabat, Libya.

Mrayati, M. (1990) ‘Statistical studies in Arabic linguistic’, in Computers and the Arabic

Language, pp.190–200, MacKay, P.A. (Ed.): Proceedings of the Arab School of Science and

Technology, Hemisphere Publishing Corp., New York.

Norman, D. and Fisher, D. (1982) ‘Why alphabetic keyboards are not easy to use: keyboard layout

doesn’t much matter’, Human Factors, Vol. 24, pp.509–519.

Pollatschek, M. Gershoni, M. and Tadday, Y. (1975) ‘Improving the Hebrew typewriter’, Technical

report, Haifa.

Wagner, M.O., Yannou, B., Kehl, S., Feillet, D. and Eggers, J. (2001) ‘Ergonomic modelling and

optimization of keyboard arrangement with an ant colony optimization algorithm’, Technical

report, Laboratoire Gnie Industriel, cole Centrale Paris, France.

Wagner, M., Yannou, B., Kehl, S., Feillet, D. and Eggers, J. (2003) ‘Ergonomic modelling and

optimization of the keyboard arrangement with an ant colony algorithm’, J. Eng. Design, June,

Vol. 14, No. 2, pp.187–208.

Yannou, B. and Hossenlopp, P. (2000) ‘Clavier alphanumérique ergonomique. Brevet d’invention’,

no 98-04588, numéro de publication 2 777 222, 23 June, Ecole Centrale Paris, Chatenay–

Malabry, France.

Zarabozo, J. (2007) ‘The miraculous Quran (part 11 of 11): a challenge for humanity’,

(IslamReligion.com), available at http://www.islamreligion.com/articles/550/#_ftnref9208.