faster on such a keyboard than on QWERTY, but found that this was not true. The key problem with an alphabetical keyboard, they concluded, was that the keys ...
Alphabetically Biased Virtual Keyboards Are Easier to Use – Layout Does Matter Shumin Zhai
Barton A Smith
IBM Almaden Research Center, San Jose, CA, USA, +1 (408) 927 1112 {zhai, basmith}@almaden.ibm.com ABSTRACT
Virtual keyboard layouts have been optimized for expert users with no consideration of novice users’ ease of locating individual keys. This paper presents a new layout produced by means of a Metropolis algorithm with an added alphabetical bias term to the previous Fitts-digraph energy function. At a small cost of expert’s performance, the new layout with alphabetical tendency offered 9% improvement to novice user’s performance. Keywords Soft, graphical keyboard; novice usability. THE PROBLEM
The need of efficiently entering text into the hand-held devices has motivated many optimized virtual keyboard layouts [e.g. 1, 2, 4]. Such virtual keyboards may allow expert users rapidly entering text by tapping keys with a stylus over minimized statistical distance. The difficulty for novice users, however, is to visually search the seemingly random keyboard and find the individual keys needed.
The energy function was used to guide a “random walk” in the virtual keyboard design space. In each step, the algorithm re-arranged the keys. Whether the new configuration was kept as the starting position for the next iteration depended on the following Metropolis function: W (A → B) = e
− ∆E
=1
kT
if ∆E > 0
(2)
if ∆E ≤ 0
where W (A-B) is the probabilty of accepting configuration B (new), ∆E is the energy change, k is a coefficient, T is “temperature”, which could be interactively adjusted. Fig. 1 shows one of layouts produced by the Metropolis methid. The potential expert performance of this layout, which depends on movement efficiency, is 42.6 wpm, assuming 4.9 bits per second IP. A period key was added to the lower right corner. Other punctuation keys can be added to the unoccupied spaces, but they are not relevant to this study.
In order to ease the novice users’ visual search difficulty, we explore an alphabetical ordering tendency to the layout without losing movement efficiency for expert users. THE SOLUTION
The most efficient virtual keyboard design to date were produced by a Metropolis algorithm [4]. The algorithm arranges the keys (“atoms”) so that the total “energy” of the “molecule” (the keyboard) was at the minimum. Energy was defined as: e=
27
27
i =1
j =1
∑∑
P ij IP
D ij Log 2 W j + 1
(1)
in which Log2(Dij/Wj + 1) was the Fitts’ law prediction of time to move a stylus from key i to key j for a given distance (Dij) and key size (Wj). IP is the Fitts’ law index of performan. Pij is the frequency of digraph ij in English text.
In Extended Abstracts of CHI2001 – ACM Conference on Human Factors in Computing Systems, Short Talks, Seattle, WA, April 1-5. p321-322
Fig. 1. An optimized layout without alphabetical ordering
To introduce an alphabetical “bias”, we revised the energy function (1) as: z
f = e + λ ∑ η ( i )( y i − x i ) ,
(3)
i= a
where e is the previous energy term defined by (1). λ is an coefficient reflecting how much alphabetical order is desired at the cost of the average movement time. η (i ) is the place of the letter i in the alphabet, with η ( a ) = −12 , η (b) = −11 , … η ( m) = 0 , η ( n ) = 1 , … and η ( z ) = 13 .
xi and
yi are the coordinates of letter i with (0.0) at the center of
the keyboard. For the space key, η (i ) is 0 at the center, and increases exponentially with distance from the center. The
result of equation (3) as an objective function was the general trend of letters starting out from the upper left corner moving towards the lower right corner. Fig. 2 shows one example of such a design, which has 41.8 wpm expert performance, only 0.8 wpm slower than the layout in Fig 1.
Subjectively, on a scale of no-preference, prefer, strongly prefer, very strongly prefer, one participant had no preference, one preferred and one strongly preferred the layout without alphabetical order. The rest of the participants either preferred (5), strongly preferred (2), or very strongly preferred (2) the layout with alphabetical order. RELATED WORK AND CONCLUSIONS
Fig 2. An optimized layout with alphabetical ordering THE EXPERIMENT
Twelve novice users participated in an order balanced within-subject experiment with the two layouts (Fig 1 & 2). With each layout, they fist tapped from a to z key as a brief warm-up. For the next 15 minutes, they entered memorable English sentences, such as “the quick brown fox jumps over the lazy dog” and “we hold these truths to be self evident. that all men are created equal.” The experiment was conducted on a Wacom PL-400 pen tablet system and an IBM ThinkPad 600. The participants, 4 female, 8 male, one left handed, 11 right handed, ranged from 13 to 43 with an average of 26 years of age. They were all fluent in reading and writing English. Results show that participants’ average speed was 9.7 wpm (words per minute) on the keyboard with alphabetical ordering and 8.9 wpm on the keyboard without alphabetical ordering. The difference (9%) between the two conditions was statistically significant (F1,11 = 6.74, p < 0.05). This difference was maintained throughout the testing session, despite the overall improvement over the trial (Fig. 3). The error rates were 2% with the alphabetical order and 2.2% without the alphabetical order (F1,11 = 0.55, p = 0.47, NS). 11
MacKenzie, Zhang, & Soukoreff [2] studied a virtual keyboard where the letters were laid alphabetically in two columns, which did not show a performance advantage, probably due its elongated shape. Lewis et al [1] proposed a 5 by 6 virtual keyboard with a strictly alphabetical sequence, which should suffer from the same problem as discovered by Norman and Fisher – the alphabetical discontinuity caused by row breaks. Instead of strictly laying out the keys in an alphabetical sequence, we introduced an alphabetical ordering tendency in the optimization process. Our layout indeed demonstrated the expected advantage for novice users. Furthermore, our approach does not suffer from the conflicting trade-off that the strictly alphabetical approach has to face – the lack of consideration for movement efficiency that will become important for expert users.
Acknowledgement We thank Michael Hunter and Jon Graham for writing the software used in the experiment. REFERENCES
1. Lewis, J. R., LaLomia, M. J., & Kennedy, P. J. (1999). Evaluation of Typing Key Layouts for Stylus Input. In Proc. The Human Factors and Ergonomics Society 43rd Annual Meeting 2. MacKenzie, I. S., Zhang, S. X., & Soukoreff, R. W. (1999). Text entry using soft keyboards. Behaviour & Information Technology, 18, 235-244.
10 ordered
9
unordered
3. Norman, D. A., & Fisher, D. (1982). Why alphabetic keyboards are not easy to use: Keyboard layout doesn't much matter. Human Factors, 24(5), 509-519.
8 7 5
The idea of using alphabetical order to ease the users’ search process on keyboards is not new. Norman and Fisher [3] studied a strictly alphabetical layout of the physical keyboard. They expected novice users to type faster on such a keyboard than on QWERTY, but found that this was not true. The key problem with an alphabetical keyboard, they concluded, was that the keys were laid out in multiple rows. The location of a key depended on the length of a row – the break point from which the next letter had to start at the left end of the keyboard again.
10
15 min
Fig. 3. Participants’ tapping speed (wpm) with and without alphabetical ordering
4. Zhai, S., Hunter, M., & Smith, B. A. (2000). The Metropolis Keyboard -- An Exploration of Quantitative Techniques for Virtual Keyboard Design. Proc. UIST'2000 - the 13th Annual ACM Symposium on User Interface Software and Technology.
