A gaze can be worth more than a thousand words with newly designed computer software that lets the eyes do the writing.
The software, called Dasher, offers greater speed, accuracy and comfort than any other system on the market, promising to transform the computing experience for those unable to use their limbs, whether due to accident or ailment, said its developers David Ward and David MacKay.
At least two computer scientists who have read the study, which appears in the Aug. 22 issue of the British journal Nature, were not ready to support Ward and MacKay's contention fully, although both praised the development as a step in the right direction for facilitating computer communication by the disabled.
The prototype tool, a refined, consumer-ready version of which will be available for free download from the Internet within a month at inference.phy.cam.ac.uk/dasher/, turns writing into a navigational task in a world painted with letters. The direction of the user's gaze controls the route followed and the characters written. Dasher uses a language model aimed at making letter selection facile and fast.
Writing tools that interpret eye movements tracked by cameras already are available, but they rely on onscreen keyboards from which users must select letters by staring at them one by one, an unnatural demand that can stress and strain the eyes, said MacKay, a physicist at the Cavendish Laboratory in Cambridge, England, and Ward, who now works for a Cambridge software company.
"When you use a standard onscreen keyboard, you have to stare at the first character you want for a specified dwell time, then stare at the next character, and so forth. This feels terribly slow and tiring. Eyes did not evolve to press buttons!" MacKay said.
To circumvent the problem, the team took advantage of the eye's natural navigating ability that routinely comes into play in such common activities as riding a bicycle or driving a car. Their design permits browsing through successive characters in a way that mimics the natural movement, making the tool more comfortable and accurate to use than an onscreen keyboard.
Users begin by looking at a letter on the screen, such as "H." The computer then pairs up the letter with fellow characters in an alphabetized array reminiscent of books arranged on a library shelf -- beginning with "HA" and ending with "HZ." The most likely combinations, such as "HA," "HE" and so on, appear larger on the screen than less probable ones, such as "HB" or "HC." This makes for easier spotting. The eyes scan the offerings on the virtual library shelf for the correct letter clusters and check out selections until the computer comes up with the complete word, in this case, "HELPFUL." It usually can do so with two letters in tow.
Given training, the software can work in most languages, including all the European tongues and Japanese.
"We aimed to make a system that would allow people to write, without a keyboard, faster than they could write before. This dream has been realized," MacKay said.
"This is an important contribution to research aimed at helping people with disabilities who have problems using traditional input technologies like a keyboard," James Landay, associate professor of computer science at the University of California, Berkeley, told United Press International. "I can't say this way is better than other ways, but it seems like a good direction."
The underlying concept is a twist on the text compression work begun in the early 1960s that aims at squeezing mounds of material into as small a space as possible.
"Our idea is the inverse of that process, turning the text compression system on its head," MacKay told UPI. "We take something that's nice and small and get a lot of text in return for the gesture -- a glance at an onscreen letter -- made."
The telling look is deciphered by an eye-tracker, a camera that, as its name implies, follows the eye to where it rests on the screen.
"The software works like a video game in which the user steers ever deeper into an enormous library," MacKay explained. "A probabilistic language model is used to shape this library in such a way that it's quick and easy to select probable sequences of characters and hard to make spelling mistakes."
The language model continues to learn so each successive use of a novel word makes it easier to write. Users can customize the system by inputting materials -- be they Jane Austen novels or scientific documents -- most representative of their writing style.
"We can't imagine a more user-friendly method for writing with one's eyes," MacKay said, comparing the method to keeping the eyes on the road while driving.
"You simply look on the screen for the first syllable you want to write. It zooms past you, and the next possible syllables appear. Most of the time, the next one you want is easy to spot, and you spot it, and it zooms by, and the next options appear. No conscious control of the eyes is needed."
The sole prerequisites for use are familiarity with the alphabetical order employed on the screen and a simple steering skill, such as necessary for driving a car, the researchers said.
Experiments with four able-bodied and two disabled volunteers, including a man with no arms, showed after one hour's practice, all could produce, hands-free, up to 25 words per minute. Users writing with other eye-tracker-based techniques employing onscreen keyboards could manage only 15 wpm after the same period, the investigators said.
"Not only is this faster than any alternative writing system driven by an eye-tracker, the frequency of spelling mistakes is about five times smaller and the new system is also less stressful to use," MacKay said.
"It is significant that they've broken past the 20-25 wpm barrier," said Jennifer Mankoff, assistant professor of computer science at the University of California at Berkeley, who specializes in computing applications for the disabled. "We're still looking for a real breakthrough on that front, but it's encouraging to see things go over 20," she said.
"I'm not sure whether the system is a significant improvement (over others)," Mankoff told UPI, "but for people using eye-trackers, there is some potential there."
Because it can be driven using any pointing device -- mouse, rollerball, touchpad or eye-tracker -- the software could have additional applications for handheld computers or mobile phones that have no space for a keyboard, MacKay said.
Compatible with any standard PC running Microsoft Windows or the open-source Linux operating system, Dasher can work in any language that can be represented by a string of symbols. As a first step towards a full Japanese version handling both Kana, a syllabic script consisting of 71 symbols, and Kanji, which uses Chinese-derived characters, Ward has written a form applicable to Hiragana, one of two written varieties of Kana.
Ward and MacKay acknowledge that Dasher is not yet perfect and needs further polishing. For example, it does not permit disabled users to send e-mail. Also, correcting errors -- which requires diverting the eyes to the left side of the screen -- can be cumbersome. The team hopes to devise a solution to the latter problem -- perhaps adding more error-correction spots on the screen -- within the next month or two.
"There are many creative ways we can communicate with each other through the computer," Mankoff said. "It's nice to see people exploring some of those creative alternatives, especially ones that can benefit those often left out of the conversation."
(Editors: More information is available at inference.phy.cam.ac.uk/dasher/images/helpful.gif.)
