By Alfred Hermida
Technology editor, BBC News website
|
The software is based on pattern recognition
|
The idea of remembering word patterns and connecting the dots might not sound like an easy way to write an e-mail.
But IBM researchers are betting that tracing letters on a touch screen will become the way to write on a handheld device like a PDA or mobile phone.
They have developed software that works by recognising the patterns of words.
"In the long run this will be one of the major interaction methods for mobile devices," said IBM researcher Shumin Zhai.
Dubbed Shark (Shorthand-Aided Rapid Keyboarding), the team say the software is ready to go into commercial development and could be bundled with PDAs in the coming months.
Palm power
Handheld computing devices like PDAs and Tablet PCs are getting increasingly powerful. But without a keyboard, taking advantage of the potential of these gadgets can be a struggle.
Anyone who has used a handheld device will be familiar with the problem.
 |
We realised that people remember patterns and not letters. There is some higher level learning that is encoded in human memory
|
Writing on screen involves tapping out words on a virtual keyboard, using software to decipher hand-writing or learning a kind of shorthand.
These methods can be cumbersome and slow, as well as being prone to errors, leading Dr Zhai at IBM's Almaden Research Center in California to wonder if there was a better way.
"The new problem is when computing takes place outside of the desktop format," he told the BBC News website.
"You wanted something that offers both high speed and high accuracy. But on the other hand people don't want to spend a lot of time to learn."
Even speech recognition software has its problems. While recording a memo works well, speech is less suited for longer, complete sentences.
"The major issue with speech is that speech is a very different process to writing," said Dr Zhai.
"Most people cannot speak an essay, an article. The way you speak and you write is very different."
Pattern recognition
Dr Zhai and his PhD student, Per-Ola Kristensson, took a sideways look at the problem and came up with some surprising results.
"We realised that people remember patterns and not letters," explained Dr Zhai. "There is some higher level learning that is encoded in human memory."
The Atomik keyboard is designed for letter associations
|
In order to tap into this unrealised potential, the IBM team came up with software that works by using geometrical patterns to represent words.
Instead of tapping out letters or writing a word on screen, you trace each letter in a single, fluid stroke. The keyboard on screen shows the shape of the word.
The software works with a standard Qwerty keyboard, as well as IBM's experimental Atomik keyboard.
The keys in the Atomik keyboard are arranged to maximise letter associations.
Dr Zhai reckons that 100 patterns cover about 40% of the words most people write.
The software lends a hand by making allowances for mistakes. It uses algorithms to recognise that some letter combinations are more likely to form words than others.
In tests, people have reached speeds of around 60 to 70 words per minute. While this is slower than touch-typing, it is much faster than tapping out words with a stylus.
"Eventually you reach a high performance level as each time you trace a word, the pattern will start to be remembered," said Dr Zhai. "It will become a recall-based skill."
An experimental version of the software has been available as a free download from IBM's Alphaworks site for a year.
The team say they are now ready to develop a commercial version that could make it way onto handheld devices in the near future.