Pen-Based Computing
An Alternative to Standard Input Methods
John Jerney
For as long as there have been handheld computers, the great challenge faced by designers has been devising an easy way to get information into one. Early palmtop devices, such as the Poqet, took the standard desktop keyboard and shrunk it down -- until it was barely usable.
The next generation of handhelds thought they had a better idea and abandoned the keyboard altogether. Led by the Apple Newton Personal Digital Assistant -- and the GO PenPoint before it -- handwriting recognition seemed like a natural for clipboard and notepad devices.
Unfortunately, early handwriting recognition systems often suffered the fate of being an impressive technology squeezed into an under-performing platform. Then along came Palm Computing. Understanding that people are often willing to modify their behavior for tangible results, the company scored an impressive success by offering Graffiti. The system promised accurate handwriting recognition if you invested a little time learning how to enter a slightly modified alphabet. To a large degree, Graffiti succeeded. But is Graffiti (and having to relearn how you make your characters) the final answer? Fortunately, the answer is no.
One future solution is accurate, speaker-independent speech recognition. When we are able to enter arbitrary text-based information by simply speaking into our handhelds, the keyboard will all but disappear. However, I believe the pen will remain, to be used as it is in real life to capture and communicate visual information such as sketches, drawings, maps, and even the occasional imaginative doodle.
Judging by the progress speech recognition has made on desktop computer systems, the dream of accurate continuous dictation and fluid command and control is close at hand. And with handheld processors reaching speeds of 200 MHz -- in the case of Digital's StrongARM 1100 and Hitachi's just-announced SH-4 -- these technologies are likely to migrate to portable systems faster than we earlier thought. In fact, some of the new windows CE Palm PCs come with voice command capability, and it is one of the central features of the new Windows CE Auto PC. (See pages 10 and 20 for more information).
But much of that capability is still the future. What about the here and now? What about the majority of devices that will perhaps never sport a processor running faster than a couple dozen megahertz, or less. And what about devices that will be designed to serve as smart cellular phones, lacking both a keyboard and a large screen. (In fact, Samsung was demonstrating just such a device at COMDEX, based on the Windows CE 2.0 operating system.)
In other words, what if you could use a standard telephone-style keypad -- with its ABC and DEF key arrangement -- to enter all the information you could ever imagine just by pressing these keys. Under this scheme, you would press each key just once, and the computer inside the device would determine which letter you meant based on how words themselves are constructed.
When it guessed wrong, which hopefully wouldn't be too often, you'd have to correct it. But under ideal circumstances, it would guess correctly most of the time.
Actually such a technology exists, and it's called T9. T9 is being developed and marketed by Tegic Communications, a 23-person company based in Seattle, Washington. To learn more about the company and its impressive technology, I turned to Don Davidge, Vice President of Sales and Marketing at Tegic.
"The Tegic founders, Martin King, Cliff Kushler, and Dale Grover, originally developed the technology to serve people with severe disabilities," says Davidge. "It was intended to serve as a communication-augmentation solution. These people typically can't write, or perhaps can't even speak."
"In particular, these people may have limited hand, head, or eye movement capabilities," says Davidge. "The idea was to create devices and software that would allow these people to input text with limited movement -- by this I mean a limited ability to create more than a few keystrokes."
One of the devices that came from this research was called 'Owl', which was an eye-tracking device hooked up to a pair of eyeglasses. "It had eight lights around a circle on which the user could focus with the gaze of an eye," explains Davidge. "Through these eight lights, the person could then initiate eight different keystrokes."
That device worked in conjunction with software somewhat similar to T9, and would convert those keystrokes to words using a linguistic database. The company soon realized that when it came to inputting text on small handheld devices, everyone was a bit handicapped by the current user interface problems. "They realized that the biggest market opportunity was in the wireless handset and handheld computing arena," recalls Davidge.
T9 uses about a dozen keys, each with three letters on it like a telephone keypad. People simply hit each key -- once -- corresponding to the key on which the letter is presented. The user then continues to type along as with a standard keyboard.
"What T9 does is decode each of the ambiguous keystrokes and instantaneously comes up with the intended word," says Davidge. "A proprietary software algorithm works behind the scenes and, in conjunction with a linguistic database, decodes the keystrokes."
"There are other systems that predict letter-by-letter -- determining the most likely letter that will come in the sequence -- but they tend not to be very accurate and they often require the user to focus a lot of attention on the screen to see if the next character is correct."
T9 instead uses a linguistic database to look at the various word options, doing so on a word-by-word basis. "What this allows the user to do is focus on the keypad itself, which is how users typically want to work with a small, portable device," says Davidge.
T9 consumes from 32 to 64 KB of memory, suitable for inclusion in just about any handheld device. It can do this since it doesn't use a dictionary. "If it was dictionary-based, it would take up far too much memory space on the device," says Davidge. "T9 is typically embedded into ROM and therefore has quite a small footprint."
"Instead of using a dictionary of, say, 60,000 words, we've encoded rules to build words. And using a high-compression system, we can encode a Ph.D. level vocabulary into a very small memory footprint."
Tegic intends to make a version of T9 available for Windows CE-based handhelds, though Davidge wasn't in a position to provide a timetable for the product's release. At least one of the new Palm PCs will include T9 capability. Another natural candidate is PalmPilot, where Davidge estimates up to only 50% of users actually use Graffiti for lengthy text input. In the meantime, Texas Instruments' new Avigo handheld organizer is the first to employ T9.
As devices continue to shrink, others are likely to include T9 capability. T9 has the right stuff for not only the new generation of smart phones, but also for quick-and-easy data entry on more conventional handhelds running Windows CE 2.0. Hopefully this will make the wait for speech recognition just a bit more bearable.
John Jerney is editor of the six-year old industry
newsletter “Pen-Based Computing: The Journal of Stylus Systems” and the
Web-exclusive monthly “mobilis: the mobile computing lifestyle magazine”
available at http://www.volksware.com/mobilis.