AN ANALYS1S OF THE STANDARD ENGLISH KEYBOARD 
Yuzuru Hiraga, Yoshihiko Ono, Yamada-Hisao 
Dept. of Information Science 
Faculty of Science, University of Tokyo 
7-3-1 Hongo, Bunkyo-ku 
Tokyo 113, Japan 
Summary 
A study of the nature of hand and finger motions in typing 
process was made, based on the data on time intervals between 
key strokes. Many of the results obtained confirm those 
formerly obtained by Dvorak and other researchers. In addition, 
it was \[bund that each key sequence is affected by its context, 
thus proving that the mental aspects of typing directly affects its 
productivity, and that the performance analysis on a particular 
keyboard arrangement is not sufficient to predict the 
performance on other arrangements. 
Based on these results, the current English keyboard is judged 
to be a less than ideal tool for typing. 
However, when carefully examined, even a rudimentary 
measurement of certain factors of typing behavior would lead to 
a good understanding of some pertinent aspects of the typing 
process. 
We faced the problem of keyboard optimization in the 
process of developing a Japanese-input keyboard. Our approach 
here is based on the measurement of the time interval of key 
strokes, obtained from the typing performance of English text 
by a proficient typist, In this paper, we first present the findings 
based on these data, discuss an important human factors 
problem involved in them, then attempt to present an 
evaluation of the QWERTY keyboard in this perspective. 
.1. Introduction ! 
\]'he task of studying typewriting behavior and evaluating 
keyboard efficiency has been taken up by many researchers in 
the past. most of these had the motivation to search for an 
optimally arranged keyboard, and have pointed out various 
defects of the current English keyboard. (called Universal, or 
QWERTYkeyboard, by the arrangement.) 
The QWERTY keyboard came into existence (both shape 
and arrangement, leaving out a few minor changes) together 
with the invention of the keyboard typewriter itself (Sholes: 
1873). Since the emerging of touch typing techniques around 
the turn of the century, it has held its established position in the 
Western society until today. But as soon as the QWERTY 
keyboard became socially prevalent, questions were raised 
against its efficiency. The ineffectiveness of QWERTY is an 
inevitable consequence of its origin. Since the most urgent 
problem that Shores faced was to avoid the jamming of the then 
mechanically deficient typebars against each other, after trying 
out various arrangements, he came out, not intentionally, with 
an arrangement difficult to type fast on. 
Well considered objections against QWERTY were voiced at 
least as early as 1893(Hammond). Since then, many studies and 
a number of suggestions for improved arrangements were made A 
Some of the ~ggestions made ~were those by HokTe(1921)," 
Griffith(1949), J Nicketls(1973),~ and M021t~1977).' Among 
them, the work done by Dvorak et.al.(1936) ,u is considered by 
many to be the most important of all, because of the deep 
psychological insight presented, and the keyboard they 
proposed, known as the Dvorak Simplified Keyboard (DSK). 
Today, DSK is thought to be of a near optimal arrangement. 
Many reports countering to these improvement suggestions in 
favor of the firmly established QW~RTY keyboard also 
appeared. The report of Strong(1956), supported by GSA, 
gave in a way a decisive blow to the DSK, and QWERTY is still 
holding its firm position. 
Typing is a highly complicated procedure deeply involving 
mental activities as well as physical movements. To cover it in 
its entirety would call for an immense task of human factors 
research, a large part of which is not yet well understood. 
2. Basic Notions 
Through our experiment, we intended to obtain some 
understanding of the relation between key sequences and the 
resulting motion of hands. In other words, we wanted to know 
what kind of key sequences result in a good typing motion, as 
against others. Since typing (especially touch typing) is a 
process of consecutive reflexive motion of two hands, it would 
not be sufficient to study this through the motion of hitting an 
individual key. The ability to tap a same key fast does not 
directly mean that the person is a good typist. The essence of 
typing skill is in the ability of maintaining minimum hand and 
finger transitions through the keying sequence, in order to 
produce good motions. Typing speed seems to be a good 
measure for this, and is also appealing to our intuition. 
Alternate measures include the fatigue of the typist, and typing 
error rate. The latter will be discussed in section 6. 
We mainly utilized our data on time intervals between key 
strokes in order to get at some aspect of hand motion. The 
elementary component that constitutes a key sequence is a key 
pair, that is, consecutive two strokes on the keyboard. Our first 
objective is to list the key pairs in the order of typing ease. A 
first order approximation for this ordering might be to put the 
key pairs in the order of typing speed. But our attempt in this 
direction immediately turned out to be futile. The time values 
themselves were not easily determinable from the measurement. 
In addition, the direct ordering gave no insight into the nature 
of hand dexterity. Sometimes it even appeared to contradict our 
intuitive notions of dexterity. 
So we turned to a two-step approach instead. We first 
attempted to factorize the time data, and next tested certain 
conjectures on typing motions based on them. 
Each of the, interval times of keying must be a function at 
least of the choice of two keys forming the key pair, as well as 
of other factors. Thus, the time. value t for a certain key pair 
(kl,k2), where suffixes 1 and 2 stand for the first and second 
keys respectively, may be described as: 
t= F(k 1, k 2) 4- e (2.1) 
--242-- 
e representing those factors that cannot be described in terms of 
the key pair, including probabilistic fluctuations. Keys may be 
subdivided according to their row, hand, and finger of operation. 
Thus, (2.1) may be rewritten as: 
t = F(hl,h2,rl,r2,fl,f 2) + e (2.2) 
where h, r, and f stand for hand, row, and finger, respectively. 
These attributes of keys may interact with each other. In an 
exact analysis, it is not correct to handle even the finger 
component of both hands together, since the keyboard is not 
left-right symmetric in shape. But we shall ignore this fact for 
the present analysis. 
A certain key may be referred to by the alphabet it stands 
for on QWERTY, or by its associated finger, row, and hand. 
Dictated by the design phylosophy of our system, we are mainly 
interested in the middle 10 keys of the upper, home, and 
bottom rows. The set of these 30 keys will be called the main 
set: Following Dvorak, some characteristic key pair patterns are 
named as follows. 2 
HurdlingStrokes by the same hand, fingers jumping over a 
number of rows. 
ReachingThe stroking of different keys with the same finger. 
Tapping.The stroking of the same key. 
Rocking:Strokes on the same hand, rolling in from the outer 
side of the keyboard inward. 
Adjacent:Stroking by adjacent fingers of the same hand. 
Remote:Stroking by remote fingers of the same hand. 
A Iternate~ 
Stroking by dill?rent hands. 
The data we use here were taken from a timing experiment 
on the Superwriter system implemented on the H-10 computer at 
the University of Tokyo. The typing was clone by a professional 
English typist (Japanese, female). Time intervals of key strokes 
were measured down to milli-seconds. Input texts were selected 
from the last part of "Alice in Wonderland" and the entire 
"Through the Looking-glass," excluding the verses ("The 
Annotated Alice", L. Carroll, ann. by M. Gardner), and 
"Information Processing" (M. Bohl), a textbook of computers. 
The whole data consists of 302,392 strokes, 142,262 for Alice 
and 160,130 for the computer text. 
Being taken from the field of literature which is unfamiliar 
to the typist, the Alice text presented to her difficulties in: 
1) The presence of special characters. The text is full of 
conversations, thus a good number of quotation marks are 
used. Also, exclamation marks, question marks appears 
frequently. 
2) Use of unusual words and phrases (a characteristic of 
Carroll). 
On the other hand, the computer ,text makes a frequent use of 
technical terms. The use of uncommon words results in a high 
rate of typing errors as we shall see in section 6. However, we 
could not conclude that these singularities would indeed affect 
the overall typing behavior differently, so both data will be 
treated combined. 
Most of the time interval values clustered in the region 
between 100 and 300 ms. The data with extraordinarily large 
values were eliminated, using a threshold of 500 ms., because 
they must have resulted from'some reasons other than the 
typing itself, such as the page turning of the text, etc. 
Of the 900 (= 30 x 30) possible key pairs in the main set, 
805 had actual data entries. Of them, the less frequently used 
key pairs showed seemingly random time-frequency 
distributions. Here we encounter our first problem, since the 
key pairs we regard of highest importance, namely, alternate 
hand stroking on the home row, had very few entries. As can 
be seen on the keyboard, these are key pairs that would seldom 
appear in a normal English text. Only one vowel, "a", is on this 
rOW. 
General Features 
1) Entries faster than 500ms. 
Entries: 294,272 Mean: 162.8 S.D.: 74.9 
2) Among l), key pairs using only the lower case. 
Entries: 180,051 Mean: 154.3 S.D,: 64.5 
3) Among 2), key pairs using only the main set. 
Entries: 179,137 Mean: 153.8 S.D.: 64.0 
3. Distribution of Individual Key Pairs 
Distribution graphs of various types of key pairs are given 
in Figures (3-I) through (3-6). Figure (3-I) is an example of 
an alternate key pair, (3-2) of an adjacent, (3-3) of a remote, 
(3-4) of a tapping, (3-5) of a reach, and (3-6) of a hurdle. Brief 
general observations of all of the key pair distributions are: 
1) Most mean values range from 100 to 200 ms. 
2) There is a lower bound for time intervals. (or, saying the 
same thing from the other side, an upper bound of typing 
speed of a key pairl) This bound falls somewhere between 
60 and 80 ms. Thus, we think that 60 ms. is the lower 
bound for keying intervals. (1,000 strokes/ minute) 
The distribution of the bound is different of that of the 
mean values, and the range is narrower. In fact, it is 
generally smaller for adjacent key pairs, though the mean 
value of their time is comparatively larger, Reaches and 
Taps, that is, key pairs that use the same finger, are 
exceptions to this. For example, key pair "d-e", a reach 
key pair, has a lower bound greater than lOOms. 
3) The peak of the graph is skewed to the left, with a long 
tail. That is, the median is seen to be always smaller than 
the mean, and the skewness factor (=normalized third 
order moment around the mean) is positive. 
4) Skewness decreases for key pairs with larger mean time 
values. It is also smaller for key pairs with less frequency 
in general. 
5) The peak is quite pointed. 
We have seen that there is a physical upper bound for typing 
speed of key pairs, which is nearly the same for different kinds 
of key pairs (except for those using the same finger). This may 
be thought of as the physical limit of response time intervals of 
fingers to the brain signals that call for separate strokes. Most 
strokes of a certain key pair gather around a time value close to 
the lower (time) bound. This means that there is constant 
rhythm in typing each key pair, or as we will see later, it may be 
better to say that nearly constant rhythm is kept within a certain 
span or conlexz, which is likely to be a word. In addition to 
these basic properties of stroking patterns, tttiere are many other 
factors which contribute to the. whole pattern of distribution. 
These fact0rsmay work in an unpredictable way, and are likely 
to have greater effects on the key pairs that are not located in 
familiar contexts. 
The effect of the frequency of usage is not so explicitly 
manifested in the distribution of individual key pairs. However, 
there exists a correlation between the mean time values and 
frequencies, which we shall see in the tbtlowing section. 
--243-- 

t = 60.5exp(-0.004t4freq) + 156.4 
which gives the statistical lower bound of 156 ms. However, 
with a frequency threshold of 50, a linear least squares fitting is 
no worse than 10% than this exponential fit in terms of the 
residue of squares, so for further analysis on frequencies, we 
shall use linear regressions. 
We further assume F in equation (4.1) to have linear 
property. The parameters for F are defined on the attributes of 
of the key pair as follows: 
h: hand transition. 
0: same hand 1: alternate hand 
r: row transition, 
Number of rows moved across in the same hand transition, 
set to 0.for alternate hand motions. 
f: finger transition. 
The distance of finger columns in the same hand, set to 0 
for alternate hand motions. 
R: row weight. 
Linear sum of weights for each row position, where the 
weights given are i, 2, and 3for home, upper, and bottom 
row, respectively. 
F: finger weight. 
Linear sum of weights on each of the finger positions of the 
key pair, where the weights given are 4.5, 4.5, 1, 2, and 3 
from the outer column inwards. 
Thus, (4,1) becomes: 
t= a+b0freq+ bl h+b2r +b3f+b4R+b5F+e (4.2) 
By multi-variate linear regression, the parameters are estimated 
to have values as follows: 
a = 185.8 
b0= -0.013 
b 1 = -40.0 
b2= 18.3 
b 3 = - 11,0 
b4= 0.514 
b5= 1.07 
These results are in agreement with our general understandings, 
that 
1) Alternate hand stroking is faster than same hand stroking, 
and hand transition is the dominant factor. The difference 
is as large as 40 ms, 
2) In same hand stroking, row transition causes a slowdown of 
around 20ms. (Row transitions and finger transitions in 
alternate hand stroking bear little meaning, and have been 
omitted from the analysis.) 
3) Finger transitions do not present a clear cut picture. In 
general, adjacent finger stroking is inferior to remote finger 
stroking in speed. 
4) Row weights and finger weights do not make noticabie 
contributions, but the orderings we chose seem reasonable. 
Further examinations were made by testing the significance of 
the difference of mean values between two types of key pairs. 
The results proved our conjectures on finger strength, the 
superiority of alternate strokes, the undesirability of awkward 
sequences, etc., with a reliability level2of over 99%. They are in 
agreement with the results of Dvorak. 
So far the analysis has been on key pairs in isolation. That 
is, we have been ignoring interactions between key pairs. But it 
turns out that these factors, namely the e factor in (4.1), cannot 
be ignored, in fact they play an important role in understanding 
the typing process, as we see next. 
5. The "Levelling" Effect 
In the preceding sections, we have stated that a typist does 
not read text by individual key pairs or characters, but by words. 
This means that she is creating a queue of elementary motions 
somewhere in her mind (not as a conscious activity), and the 
actual finger motions are generated as an aggregate of the 
components in this queue. From this we infer that a certain 
optimization of the motions of hands and fingers takes place in 
this process. 
The analysis in the preceding section was concerned with 
isolated key pairs and ignored interactions among them. 
Because of this simplification, the multi-variate model enabled 
us to explain only half of the residue of the time-frequency 
regression analysis. The, rest must be a result caused by external 
effects, which is represented by the e factor in (4.2). So, we 
proceed to examine the effect of the context keying pairs to the 
time values of a key pair. 
We mean by context the key pairs immediately preceding or 
succeeding the key pair under attention. A simple tallying 
showed that in convex situations (that is, a situation that the 
preceding and succeeding time values are both smaller than the 
present time value), and in concave situations (the opposite of 
the convex), the time value tends to be pulled towards the 
direction to relax the curvature, with respect to the mean time 
value of the present key pair. The effect is most significant with 
alternate hand stroking, where 87% of the time values in a 
convex situation are pulled downwards, and 59% in a concave 
situation are pushed upwards. This means that the context is 
more effective to increase the overall speed rather than to slow 
it down, though this is the opposite to our initial expectation. 
At any rate, our hunch that the time interval of consecutive key 
pairs are smoothed, or levelled, seems correct. 
A more detailed analysis confirmed the qualitative claim 
made above. Naming the preceding, present, and succeeding 
time values tl, tg, and t 3, respectively, the correlation of 
T 1 = (t2-0/ff- 
T 2 = ((t I +t3)/2-t)/s 
was calculated, where t is the mean, and s is the standard 
deviation of the present key pair. T 1 has a (0,1) distribution. 
The distribution of T 2 is unknown, although it is in a sense 
normalized with respect to T 1. Counting all the cases, the 
correlation coefficient of T 1 and T is 0 4325 while for convex 
2 ' ' situations this is 0.6701 and concave 0.5910. The concave one 
improves to 0.6282 for alternate stroking, while convex-alternate 
is 0.6479 in this case. These values of coefficients suggest a 
strong positive correlation between T l and T2, especially for the 
convex and the concave, or popped out cases. The result that 
alternate hand stroking is more affected by the context means 
that the time intervals of these key pairs have a flexible nature. 
The existence of this flexibility is quite obvious, since ~,-actically 
no factor restrains these key pairs, and the two strokes are quite 
independent of each other. 
We may view the outcome here from the point of typing 
rhythm. All indications we have are in support of a view that 
such a levelling process is inherent in good typing behavior. In 
other words, there is a clear tendency to work towards a 
constant rhythm. It is hard to evaluiate this effect quantitively, 
but we believe this plays an important role in the whole typing 
process. 
--245-- 
Since this levelling must result from some mental 
scheduling procedure, it is likely that the more skilled a typist 
becomes, the more she will work towards rhythmical typing. 
This would mask the inherent weakness of the keyboard, as 
sugar coating covers up the bitterness of tablets. Therefore, the 
results of a superficial analysis of the performance on the 
results of a given keyboard is not directly usable to predict the 
expected performance on an entirely new keyboard. 
6. T.vpin~ Errors. 
Errors occur quite frequently in typing, as was the case with 
our experiment. Here, we will attempt an analysis of the nature 
of common errors. The error statistics we obtained are given in 
tables (6-1) and (6-2). 
Typing is a complex process involving many levels of 
human intellect and motion, and errors may occur in any one of 
these levels in various forms. For example, if the typist learns a 
word in a misspelled form, she will make constant errors with 
this word without even noticing them. But these errors, which 
result from visual or intellectual reasons, are not the kind we 
are interested in here. We shall investigate those that occur in 
conjunction with the mechanical motions of hands. These 
errors may occur either because the motion itself is inherently 
prone to errors, or because the typist has a particular deficiency 
for that motion. An example for the latter case is seen in our 
subject who has troubles with the word "little", typing it as 
"litle", "liitle", "littl", "litl", et cetra. 
Generally, the pattern of errors depends on individual 
typists (much more than the distribution of typing speed does) 
and since we had only one subject in the experiment, the 
findings given below must not be thought of as a general result, 
though it should reveal at least a part of the truth. An 
additional limitation is that the subject was not accustomed to 
the keyboard used in the experiment. The effects of such 
factors as the reactive force of the keys against fingers, or the 
arrangement of special characters, may not be ignored. 
Table (6-1): Errors in the Main Set. 
Total 897 instances (602 distinct) 
Omission 65.2% 
Insertion 4.7% 
Replacement 2.8% 
Interchange 2.2% 
Chatter 22.5%(Overlapped) 
Other errors 17.6% 
Table (6-2): Distribution of Omission 
Errors. (The middle key of the sequence 
is skipped.) 
Text sequence 
L-R-L 7.4% I R-L'-R 2.6% .... 10.0% 
L-L-R 4.3%. 
R~-L 4.8%J... 9.1% 
L-R-R lO.5Z. 
R-L---L 30.2% j''" 40.7% 
L-L-L 6.9%. 
R-R-R 2.9%J... 9.8% 
Here, we will not look into errors with shift keys, although 
shift key errors form a category of their own. They are likely to 
occur in an interrupted state of mind, that is, when the mind is 
conscious of the typing behavior. The mistouching of either the 
shift or the character key that occur here is of a different quality 
from other simple mistouching, and these errors are not a good 
example of a deficiency in continuous reflexive motion. One 
typical example is that the typist went on typing a few lines 
without noticing that the shift key was locked. 
The types of errors that occur within the main set are fairly 
limited. They are categorized as follows. 
1) Omission: Errors that skip characters of the text. This is 
the most common type of error that is made by a well 
trained typist, since her touch, is lighter than that of the 
less trained. They will be occasionally too light to go over 
the threshold of the key. These errors are found more 
frequently in faster sequences. 
2) Insertion: Error that inserts extra characters that are not in 
the text. 
3) Replacement: Error that hits an irrelevant key in place of a 
proper one. The most likely key to be mishit is the key 
adjacent to the proper one. 
4) Interchanging: Error that hits a key pair in the reverse 
order. This appears to be a result of the peculiarity of 
individual typists. 
5) Chattering: Error that a certain key is doubly typed. This 
error is characteristic of some electric typewriters. It often 
occurs together with the omission of the next character, 
which indicates that it occurs when the second key is hit 
while the first is still being depressed. The chattering itself 
may be due to too sensitively made keys, or to the 
trembling of fingers, or both. 
The occurrences of errors are not evenly distributed, but 
are clustered. This implies that errors are in part due to some 
mental and physical state of the typist, especially the fatigue. At 
the same time, we see that a same error is repeatedly made, 
which means that the errors are not entirely due to carelessness, 
or statistical freaks. Errors are made more frequently by the left 
hand, although it may not be correct to infer this to be an 
indication of the inferiority of the left hand, since the total 
number of strokes typed by the left is also greater. 
In the data, we see that more than 60% of the errors are 
omission, errors. Furtiler inspection shows that about 40% of 
the letter skipping takes place where the text goes either 
L-R-R-, or R-L-L-, where the s~cond stroke in the sequence is 
the skipped key. From The previously obtained results, we 
know that alternate hand pairs can be stroked faster than same 
hand pairs, so we may infer that, because of the levelling effect, 
the typist would unconsciously try to speed up the same hand 
pair and such psychological stress would result in the skipping of 
the first key of the pair. Errors are made also where the 
sequence turns from an alternate hand mode into a same hand 
mode, which should accompany a similar psychological stress. 
The results obtained here are consistent with the observations in 
the preceding sections. 
Errors tend to occur more frequently with words which are 
less familiar to the typist. A typical example is seen in the word 
"magnet" and their derivatives. The keying sequence of this 
word itself does not contain highly awkward sequences. 
Nevertheless, there are a large number of errors associated with 
them, so words which are unfamiliar to the typist must be more 
prone to errors. This aspect cannot be inferred from our 
discussion concerning the abstract natures of motions, so that it 
should be explained through a depth analysis of the particular 
motions associated with the word, in terms of psychological 
--246-- 
hesitation, analogy to familiar and similar words, lack of pattern 
practice for the entire word, etc. 
7. Further Discussions 
Briefly summarizing the observed results, we find that most 
of the conjectures stated about the dexterity of hand and finger 
motion is true for individual key pairs. However, the effect of 
these factors are not as pronounced as we first anticipated in the 
data, and taken as a whole, the data seemed to be quite 
dispersed. We attribute this to the effect of the context key 
pairs that work on them, and surmise that to think of a keying 
sequence merely as a juxtaposition of individual key pairs would 
not be sufficient, even at the physical level. 
At the mental level, it is well known that typing is not a 
collection of individual letter typing but a typing of a longer 
pattern as a unit. A typist takes commonly used syllables (e.g. 
"-tion", "-ing", "-tive", etc.), words, or even phrases as a unit of 
recognition, not individual key pairs or characters. In the 
recognition, the text is segmented according to whether or not 
the typist has the knowledge of a particular chunk. This should 
be very closely related to the reading process in general. In 
typing, the term "knowledge" does not mean only that the typist 
understands the particular word, but also that she has a 
canalization, or a chunking ability, in her mind that translates 
the word into a sequence of keying motions. If the typist does 
not "know" the word, then it must be interpreted by syllables or 
maybe characters. Better defined canalization exists for 
frequently used words, hence for frequently used key pairs. 
What we find in the results of our experiment is that this 
canalization exists not only in the mental aspect of typing, but 
also in actual physical motions of hands and fingers in typing. 
As a result, if a keying sequence contains a larger proportion of 
key pairs which are harder to execute, then even easier key pairs 
are slowed down in spite of the potential speed-up due to 
learning, and the overall typing speed settles down at a value 
which is below the attainable level as an aggregate of individual 
key pair performances. This is perhaps the main reason for the 
discrepancy between the actual reported performance figures on 
Dvorak keyboard, and the figures predicted for it by various 
authors based on the performance on QWERTY keyboard. 
8. An Evaluation of the QWERTY Keyboard 
What we have been observing in the preceding sections may 
be directly related to the evaluation of the effectiveness of the 
QWERTY keyboard. 
Through the tallying of the text, we have the distribution of 
keystrokes on QWERTY as in table (8-l). The figures tell us 
that the left hand is overloaded, there are too many strokes 
typed on the upper row, the fingers are loaded poorly in relation 
to their dexterity, etc., all of which show the weakness of the 
QWERTY keyboard. Furthermore, the rate of awkward 
sequences is strikingly high; Of all the key pairs in the text, 
9.6% are hurdles and 8.2% are reaches. 
These results in themselves indicate the deficiency of the 
QWERTY keyboard, but these are not the crucial points. From 
the measurements made by others in the past, we have been 
told that for all these deficiencies, the QWERTY keyboard is no 
worse than by 20% in typing speed than by DSK. Some 
reported that the diIti~rence between DSK and QWERTY 
Table (8-1): Stroke Distributions. 
Hand Distribution (Shift key not counted. 
Left 48.0% 
Right 35.9% 
Space bar 16.1% 
Row Distribution (Space bar excluded.) 
Top 0.1% 
Upper 51.5% 
Home 31.8% 
Bottom 17.6% 
Finger Distribution (per hand) 
Left Right 
Index 37.2% 45.3% 
Middle 34.8% 21.4% 
Ring 14.0% 27.6% 
Little 14.1% 5.7% 
predicted through the analysis of QWERTY data is within °nly a 
few percents. Why can the diffel:ence be estimated so small? 
We think that this is due to the levelling effect, in which an 
expert executes hand. motions through optimal paths. 
Supposedly slower stroking sequences will be pulled to be faster 
(and possibly become more prone to errors), and yet the faster 
, ones might be made slower. The difficulty that is inherent in 
the keying sequence of the text on QWERTY, with all the 
awkward sequences and incessant wasted motions, is 
transformed into the tension of the hands of the typist in order 
to compensate for it, in an effect to type as rhythmically as 
possible. In this way, the resulting speed may not be aft~cted as 
much as should be expected from the difficulty of individual 
strokes. Thus, when used to predict performance on DSK, such 
data will not give us a good estimate. The existence of 
unpreferred keying sequences affects the local performance of 
keying. In addition, continuous such effort on QWERTY 
certainly causes the fatigue of the typist, and it might even 
become hazardous to her health. 
Based on these observations, we must reject the popularly 
stated notion that the QWERTY keyboard is a fully satifactory 
arrangement \[br English texts for most purposes. 
9, Concluding Remarks 
Many of the results we obtained confirm formerly stated 
claims on a basis of experimental results. The experimental 
proofs for the fact that the context of the text directly affects the 
production of the typing should be of value, for two reasons. 
One is that it shows that typing is a highly mental process, as 
well as a physical one of moving hands and fingers. Two, and 
what we wish to emphasize, is that this effect works in the 
direction of relaxing the tension that might occur in the typing 
process, so if we look at the typing behavior superficially, we 
might arrive at wrong conclusions about hand dexterity and the 
effectiveness of keyboards~ oReports which support the efficacy 
of the QWERTY keyboard v'" appear to be making this error. 
--247-- 
Acknowledgement: We would like to thank Dr. S. Kawai for 
helpful suggestions and discussions on our work, and would also 
like to express our gratitude to Mr. M. Mogaki and Mr. J. Jan 
for helpful comments. 

References 

Yamada, Hisao; "A Historical Study of Typewriters and 
Typing Methods: from the Position of Planning Japanese 
Parallels" (February 1980) Journal of Intbrmation 
Processing, Vol. 2, No. 4, pp.175-202. 

Dvorak, August; Merrick, Nellie L.; Dealy, William L. & 
Ford, Gertrude C.; "Typing Behavior, Psychology Applied 
to Teaching and Learning Typewriting" (1936) American 
Book Co., New York 521pp. 

Dvorak, August & Dealy, William L.; "Typewriter 
Keyboard" (May 1936) U.S. Patent 2,040,248, 8pp. 

Hoke, Roy Edward; "The Improvement of Speed and 
Accuracy in Typing" (1921) Ph.D. Dissertation, Johns 
Hopkins University, Baltimore, 50pp. 

\[5\] Griffith Roy T.; "The Minimotion Keyboard" (November 
1949) Journal of Franklin Institute, Vol. 248, 5, pp.399-436 

\[6\] Nickells, Robert F. Jr.; "The Design of an Optimal 
Typewriter-like Keyboard" (1973) M.S. Thesis in Industrial 
Engineering, Lehigh University, Bethlehem, Pennsylvania, 
vii+57pp. 

\[7\] Malt, Lillian G.; "Keyboard Design in the Electric Era" 
(September 1977) Conference Papers on Developments in 
Data Capture and Photocomposition, PIRA Eurotype 
Forum, September 14-15, London, 8pp. 

\[8\] Strong, Earl P.; "A Comparative Experiment in Simplified 
Keyboard Retaining and Standard Keyboard Supplementary 
Training" (1956) General Services Administration, 
Washington D.C., vi+42pp. 

\[91 Kinkead, Robin; "Typing Speed, Keying. Rates, and 
Optimal Keyboard Layouts" (October 1975) Proc. of 19th 
Annual Meeting of Human Factors Society, Dallas, Texas, 
pp.159-161 
