Shark (Shorthand-Aided Rapid Keyboarding) is an advanced pen-based shorthand method that allows users to input words into mobile devices by tracing them letter by letter on a virtual keyboard. Instead of tapping independently on four virtual keys with a stylus to spell "word," for example, consumers would put the stylus on "w" and then carve a continuous trail all the way to "d."
"What we are doing is creating a human-machine code," said Shumin Zhai, a research staff member at IBM Almaden Research Center, said during a presentation at the New Paradigms for Using Computers conference in Silicon Valley on Monday. "It uses geometrical patterns to represent words." As an aside, he added, "We're back to carving symbols on rocks."
Users initially hunt for letters to write words, but the idea is that they fairly rapidly start to memorize the shape of common words and word components--and therefore, their dependence on visual guidance decreases. The computer assesses the user's final pattern, interprets it as a word from its database and turns it into text on the screen.
Shark, which has been in beta since October 2004, remains a lab project, but it shows promise, said Zhai, who added that after four training sessions, users were able to memorize about 60 patterns. "That covers about 40 percent of the words (and word components) that people commonly employ," Zhai maintains.
The growing popularity of smart phones is prompting researchers and companies to develop input systems that will work optimally with those devices.
"Laptops are not really mobile devices," said Ian Smith, a researcher at Intel Labs in Seattle. "They are effectively nomadic devices."
Consumers have adapted to the standard QWERTY keyboard, but these keyboards are generally too large to tote around with mobile devices. Some companies have come out withShark speaks English well that project a laser image of a QWERTY board. The keyboards are light, but require a flat, even surface, said Zhai--not always easy to find on a bumpy subway ride.
Speech and handwriting recognition have been tried for mobile devices. Speech, however, requires people to think in complete, articulate sentences, a cognitive trick that may have faded out with the orators of the Roman empire. And handwriting is slow, Smith added. Smaller keyboards such as the take time to learn to navigate.
Shark, according to IBM, solves many of those problems. Some users have gotten around 60 to 70 words per minute with the program, and while that's slower than touch typing, consider how many words you can get per minute with a Palm stylus today.
The Shark keyboard system also tends to be compatible with natural English language patterns. The average word is 4.7 letters long, he said, which means that the system doesn't get overtaxed by elaborate scribbles. While that means there are 11.8 million five-letter combination, the actual number of combinations people use is far lower. The average person has a writing vocabulary of about 10,000 words, so the patterns the computer will have to recognize remains somewhat contained.
Shark works with a variety of keyboard types, and IBM is currently experimenting with a number of them. QWERTY keyboards, which were arranged to prevent mechanical problems caused by poorly configured keys, don't work that well. "The shapes become more complex than necessary," he said.
Alphabetically arranged keyboards are slightly worse, he added, because the distance between some commonly associated letters is even farther apart.
Instead, IBM is conducting experiments with a modified version of its Atomik keyboard. In that keyboard--which consists of three six-letter keys sandwiched by an upper and lower row of four letter keys--keys are arranged to maximize letter associations. The first Atomik board was designed for touch typing, but the slightly modified version appears to work well with Shark.
Atomik, though, isn't perfect: "ING" and "ED" verb endings are spaced less efficiently than IBM would like.
Errors in writing can be corrected with a drop-down menu. IBM is also determining how much leeway to give users in making their lines.
Beta copies of Shark can be downloaded from IBM's AlphaWorks site, where it releases experimental software.
Others at the conference explored different options for getting around the tiny keyboards present on most mobile devices. Jeff Pierce, from Georgia Tech's college of computing, for example, talked about how such devices can take over nearby resources, such as a laptop keyboard or a desktop monitor.