A Method for Optimizing Single-finger Keyboards

The physical arrangement of characters on keyboards contributes to the efficiency and rate at which messages can be generated. While this is true for any individual and any keyboard, character arrangement can have a particularly profound impact on persons with disabilities who utilize only a single finger (or headstick, mouthstick, laser pointer, etc.) for key selection. We propose a new method for optimizing character arrangement on a fixed set of keys, based loosely on the pioneering work of Levine and Goodenough-Trepagnier. By repeatedly swapping the positions of different character pairs (or triplets) and re-evaluating the efficiency of the arrangement, our technique rapidly converges on the most efficient character layout. This method is applicable to any keyboard and can be tailored to the specific motor tendencies of a particular user or class of users. BACKGROUND Researchers in augmentative communication have been concerned with efficient keyboard layout for many years [1, 2]. The optimization of character arrangements has recently become important to a broader population as small electronic devices become more prevalent. The arrangement of the traditional qwerty keyboard is (very roughly) optimized for ten-finger typing. On smaller keyboards, where a single finger or stylus is used for typing, this arrangement is very inefficient – it takes a great deal of extraneous movement to generate text. Frequent keys like "a" and "o" are near the edge of the keyboard and common two-letter sequences such as "sh", "sp", and "cl" require selection of keys on opposite sides of the keyboard. Of course, the same inefficiencies confront an augmented communicator who types with a single finger, a headstick, or any other pointing device. Levine and Goodenough-Trepagnier [1] describe a general methodology to optimize keyboard arrangements for users with specific motor impairments. They focus solely on minimizing the total motor cost associated with using a given arrangement to generate text. Of course, the motor cost of a keyboard is only one consideration in designing a character arrangement appropriate for a particular user. One must also consider cognitive factors, such as the ability of the user to memorize the layout and the intuitiveness of the arrangement. Nevertheless, minimizing the motor cost can represent an important step in the keyboard design process. We have previously described a method for optimizing the arrangement of letters on an ambiguous keypad which has multiple characters on each key [2] – an interface also addressed in [1]. We have generalized the technique used in our earlier study to optimize character placement on standard (one character per key) keyboards. This rapid optimization procedure reliably produces the minimum cost arrangement for a given set of constraints. METHOD We have adopted a simple but effective measure of the motor cost of a given keyboard. First, one defines a motor cost associated with moving from a given key to any other key [1]. This cost is based solely on the keyboard layout and the motor abilities of the user. For an able-bodied person and keys of a fixed size, these key transition costs might just reflect distance between the centers of Optimization of Single-Finger Keyboards the keys. For a person with specific motor deficits, or for keyboards with keys of different sizes, a more appropriate set of transition costs would be used. For example, Levine and GoodenoughTrepagnier propose a model in which transition costs are lower along a preferred elliptical axis [1]. Given a set of transition costs, the total motor cost necessary to produce a representative text can be approximated by summing the cost of each key-to-key transition multiplied by the number of times that transition appeared in the text. To determine the frequency of character-to-character transition, each unique two-character sequence – called a bigram – in the reference text must be counted. In our studies, we have derived bigram statistics from a 3 million word text. Given a set of bigram frequencies and key-to-key transition costs, we attempt to minimize the total motor cost required to reproduce the reference text. Our optimization scheme is based on earlier work done by our research team for the optimization of ambiguous keypads (like on a telephone) [2], which in turn was inspired by research in the field of operations research [3]. In this approach, characters are initially assigned to random keys. The total motor cost of this arrangement is computed as described above. We then try to decrease the motor cost by systematically re-arranging the characters. This is accomplished by repeatedly selecting pairs of characters and computing whether the cost would increase or decrease if their positions were swapped. The characters are actually swapped only if the cost decreases. Because most of the characters remain in the same position during each rearrangement, determining the relative cost of a two character swap can be done very rapidly by the computer. After repeating the swapping process many times, the character arrangement will eventually stabilize such that there are no advantageous character swaps remaining. This arrangement may not have the absolute minimum total motor cost, but it will have the best that is possible using the swapping technique given the initial random arrangement. To reduce the impact of the initial arrangement, the optimization procedure is repeated many times with different starting keyboard layouts. Only the final arrangement with the lower total motor cost is retained. RESULTS We applied our optimization techniques to a traditional keyboard layout and a modified 6 by 5 grid with enlarged space keys. The motor cost associated with moving between any two keys was assumed to be proportional to the distance between the key centers. The resulting character arrangements, produced in less than one minute on a 300 MHz computer, are depicted below. Compared to a qwerty arrangement, the optimized arrangement on the traditional keyboard reduces the total motor cost by 35.8%. Compared to an alphabetic 6 by 5 grid layout, the optimized grid arrangement reduces the total motor cost by 32.0%. When key-to-key transition costs are biased to reflect specific motor tendencies, even greater gains are possible. For example, if horizontal movements are twice as "expensive" as horizontal movements, the optimized traditional keyboard reduces the total motor cost by more than 40% over a qwerty arrangement.