keyboard, or word-gesture keyboard in this article. It is
a re-imagination of the conventional key striking-based
keyboard. The paradigm have been also known as
shorthand-aided rapid keyboarding (SHARK), 21, 46 shape
writer or shape writing, 23, 33, 47 and can also be called gesture, graph, stroke, trace, swipe, sweep, slide, or glide
keyboard. This paradigm has not only been extensively
researched in the academic literature15, 17, 19, 21, 23, 46, 51 but has
also already been embodied in many products. To date,
different implementations of this novel paradigm have
been marketed by a number of companies under at least
the following brands: ShapeWriter, SlideIT, Swype, T9
Trace, Flex T9, and TouchPal on a great number of devices.
This article summarizes a decade-long academic research
that led to the establishment of this input paradigm.
We developed the basic concepts and initial prototype
of a word-gesture keyboard from 2000 to 2002, 16, 46 and
took many more years to mature and deploy the technology. 21, 23, 50, 47, 51 The paradigm itself is still emerging and
developing, with both necessity and opportunity for further
technological advances and better user behavior and performance understanding. We outline some of the future
research directions at the end of this article.
2. theoRy, RationaLe, anD DeSiGn PRinCiPLeS
oF WoRD-GeStuRe KeyBoaRDS
The basic type of input action on a traditional keyboard
is striking an individual key. To do this well requires good
tactile feedback. On a touch screen, another type of input
action is possible. Instead of a striking action, one can use
a continuous stroke gesture to convey information. Indeed,
it is compelling to use sliding gestures on a touch keyboard
for functions such as DELETE or SHIFT. 3 In early 1980’s,
Montgomery32 conceived the idea of using sliding gestures
on a touch keyboard to enter characters. He designed a “wipe
activated” keyboard with a flat touch sensitive surface. The
positions of the letter keys were carefully arranged to make
consecutive letters commonly appear in words connected
on the keyboard. The user can slide across adjacent letters
to enter a string of letters. Montgomery believed such continuous “wiping” actions are more efficient hence “bringing
manual input into the 20th century” from 1860’s Qwerty keyboard. Perhaps ahead of its time, Montgomery's pioneering
work had very limited impact, with only a few citations in the
literature. Without further research or actual deployment, it
was also unclear how easy or efficient it was to use such a keyboard which required detecting or remembering connected
sequences of letters in order to wipe through them.
Stemmed from our work on optimizing stylus tapping
keyboard, 48 we envisioned the paradigm of word shorthand
gesture keyboard for touchscreen devices. On a word-gesture
keyboard, instead of tapping individual keys or wiping
through a sequence of letters connected on the keyboard,
the user can write each and every word in a lexicon via a word
gesture (also referred to as sokgraph—short hand on keyboard as a graph50). A word gesture approximately traces all
letters in the intended word, regardless if they are adjacent.
For example, to write the word fun a user touches the f key,
slides to the u key then the n key, and lifts up. The resulting
gesture is analyzed by a statistical model and the most likely
word (in this case fun) is selected and entered by the system,
which optionally also displays alternative N-best candidate
words (Figure 1).
Note that the meaning of “word” in a word-gesture keyboard lexicon is broadly defined. While most words can be
selected from a natural language, some can also be tokens
defined by arbitrary strings of characters, such as gmail.
com. Each such token in turn defines a word-gesture, or a
token path, on the keyboard.
2. 1. Gesture keyboard feasibility
The first question that may arise here is why the word-gesture keyboard paradigm is possible at all, considering that most word gestures will run across letters that
are not part of the word intended. Indeed, this challenge
seemed to have prevented the attempt by Montgomery in
the early 1980s32 toward establishing such a paradigm.
Montgomery32 instead proposed to rearrange the keys to
maximize the chance for a user to be able to wipe through a
sequence of adjacent letters that happen to make a word or
a common word fragment without lifting.
However, this problem was not insurmountable. As
Shannon36 observed and elegantly demonstrated in his classic paper on information theory long ago, there are strong
statistical regularities in natural languages. For example,
Figure 1. Shape Writer on the iPhone is an example of a word-gesture