| It'll be a long time before handwriting recognizers are
as good as pharmacists at interpreting anybody's sloppy
scrawl. Therefore, the designers of today's PDAs must
compromise the ideal of always recognizing everything a
user writes. One alternative is to minimize freehand pen
input by offering selections in menus. Another is to
capture an image of the user's handwriting without
attempting the difficult translation into ASCII text. The latter approach is known as digital ink. Some PDA designers contend that digital ink is not a compromise or a workaround. For certain applications, they say, ink is an entirely appropriate data type. Often-cited examples include brief reminder notes and entries in appointment calendars. In neither case, they argue, is a user likely to need full-text searching or sorting, the main advantages of ASCII text. "There's an underlying assumption that 'before I can do useful work, it's gotta be in ASCII,'" notes Gregory Stikeleather, president of Aha Software (Mountain View, CA), which makes an ink-oriented word processor called InkWriter. "That isn't always the case." Still, it's hard to believe that universal recognition wouldn't be implemented if it were reliable, adaptable, and didn't tax a PDA's resources. For one thing, byte-coded text requires less storage space than digital ink — an important consideration for hand-held devices with small RAM-based file systems. And there's always the chance that a user might indeed want to search a series of reminder notes for a key phrase or look for a particular name in a large appointment book. For the immediate future, though, digital ink will be an integral feature of PDAs, and designers are taking different approaches to it. These differences are typified by Apple's MessagePad and Tandy/Casio's Zoomer. Both of these PDAs let you tap an icon to switch the handwriting recognizers off. This lets you write freely in digital ink, which the computer captures and stores as a graphical image. This image is not a simple bit map of pixels, however. Both PDAs save your writing as vectorized strokes, similar to the difference between bit-map fonts and outline fonts. When compressed, the strokes require less storage space than bit-map images and also preserve information that could be useful later (see the text box "Jot Defines Electronic Ink" on page 110). The Zoomer makes the best use of this information. It supports deferred recognition, the ability to translate ink into text anytime after it's captured. You merely select the ink, turn on the recognizer, and confirm that you want to translate the writing into text. This is where saving penstrokes becomes important. It's much easier for a recognizer to interpret stroke data instead of a plain bit map, because the strokes preserve the way in which a character is written, not just its image. If nothing more than a bit map were saved, the recognizer would be reduced to optical character recognition, which is not very reliable for handwriting. The MessagePad doesn't support deferred recognition, although nothing in the Newton Recognition Architecture precludes it from being added in the future. Apple gives two reasons for this decision: Real-time recognition saves storage space, and usability testing has showed that people tend to write much sloppier when the recognizer is off, making it more difficult to recognize. As a result, the MessagePad's approach to digital ink is somewhat less flexible than the Zoomer's. In truth, however, the Zoomer probably needs digital ink more than the MessagePad. The Zoomer's recognition is noticeably slower (a consequence of the Zoomer's 8088-compatible CPU versus the MessagePad's ARM610 RISC chip); it's less adaptive (it recognizes only printing, not cursive writing, and it isn't trainable); and it is somewhat less accurate (based on results of preliminary tests). A key difference between the two recognizers is that the Zoomer interprets words character by character, while the MessagePad lets you choose between a dictionary-based system, character-by-character recognition, or a combination of the two. Normally, the MessagePad tries to find each word you write in its 10,000-word dictionary. If it can't find a match, and if the character-by-character option is off, the MessagePad makes a guess — often a very wild guess. For example, the MessagePad stumbled badly on the proper name Halfhill, even though both parts of the compound word were in its dictionary. Five attempts yielded these five guesses: florists, teachers, forecasts, four their, and Clarence. One solution to the problem is to add the word to the dictionary by spelling it on a pop-up QWERTY keyboard. Once I added Halfhill, the MessagePad got it right every time. The other alternative is to activate the character-by-character option, which works fairly well but slows down the recognizer and reduces overall accuracy. The Zoomer's answer is to always offer the option of digital ink — unlike the MessagePad, which offers it only in certain applications. "If I were writing poetry, a dictionary-based recognizer would be great," comments Jeff Hawkins, founder and chairman of Palm Computing. "But when you start writing names and other words that aren't in the dictionary, you start running into problems." Copyright 1994-1997 BYTE |