The Word-Gesture Keyboard: Reimagining Keyboard Interaction

In this paper, Zhai and Kristensson discuss their Word-Gesture Keyboard which allows users to drag their finger in a single stroke across a touch screen keyboard to input a word. Their keyboard uses a probabilistic model to provide a possible-word menu from which the user can select their desired input.

Zhai and Kristensson first describe the requirements for the widespread acceptance of a new text input method:

1. High Speed of Input
2. Minimal Cognitive Load
3. Support for the Development of Efficiency

From there, the researchers discussed the feasibility of a gesture based text input method. They note that English, like nearly all natural languages, feature a high degree of regularity and redundancy and that the average word length is only about 4 to 5 characters. Longer more complex words can still be entered through the striking method so all words can be entered by users despite their particular skill or accuracy in using the keyboard. Essentially, SHARK (short-hand aided rapid keyboarding) is really an improvement of the existing keyboard. The probabilistic recognition of matching words in the English lexicon also reduces the impact of inaccurate input caused by the fat finger problem.

Since the user doesn't have to lift their finger, the input is generally takes less time than tapping.
Their keyboard input method does not require any training which is a must for any new text input method. Since most input is done for a small set of commonly used words, these common words' input paths are easily remembered through procedural/muscle memory which further improves input speed. This training allows users to progressively become more efficient which is a positive mark of a  "progressive user interface."

Their keyboard also has features common system commands...

• to change the case of the word before the caret to further decrease time to input and minimize user frustration.
• to copy and paste
• to switch languages (lexicon search and different character input sets)

Zhai and Kristensson performed several experiments to test...

• recall rates without visual feedback (looking at the keyboard) across multiple sessions
• resulting in about 13 words remembered for the 1st session to 45 to 57 words remembered for the 4th session
• entry rates (speed) for initial performance with varying amounts of time to practice on QWERTY and ATOMIC keyboards
• resulting in 7 to 13 wpm for ATOMIC and 15-25 wpm for QWERTY
• maximum performance for small set of text inputs recorded in the form of a competition
• with an average error-free performance of 57.5 wpm and top performance of 99 wpm

A subjective evaluation was also conducted with responses and comments gathered from their official release with the Apple App Store for the iPhone. Responses were 81.6% completely positive, 12.5% somewhat positive, and 5.9% completely negative.

Given the limited input space touch screens for mobile devices, it is impossible to use the standard two handed typing for which the DVORAK keyboard was designed. As such it makes sense to explore less traditional methods of keyboard input and to attempt to ease input for these devices, especially in light of their widespread use. Zhai and Kristensson's work is a great work that does just that with the grounded scientific work and forethought given to it that such work deserves.


Shumin Zhai and Per Ola Kristensson. 2012. The word-gesture keyboard: reimagining keyboard interaction. Commun. ACM 55, 9 (September 2012), 91-101. DOI=10.1145/2330667.2330689 http://doi.acm.org/10.1145/2330667.2330689