There Be PDFs Ahead!

Anyone who was reading technology sites back in spring of 1999 might remember an alternative to Palm, Inc.'s Graffiti that was making headlines.  Called Quikwriting, the system was designed so that a user never had to lift the stylus between entering characters, and it was suggested that it would provide faster input than the unistrokes-based system.  For those of you who weren't around, here's some links to articles: Wired, Slashdot, and the original short paper on it from the 1998 UIST symposium.

There was tons of anecdotal testimony kicking around, but no actual results that adressed the two main factors: was Quikwriting in fact faster, and just how much training was required for these speeds to be reached?  (The Wired article makes reference to formal tests being done, but I've never managed to find them, or seen them referenced in other papers.)  While, personally, I wouldn't dispute whether it was in fact faster—the training bit, and therefore its immediate usability, was always something I wondered about.  One of the nice things about Graffiti was that tests have shown that not only is it easy to learn, it's also easy to remember.  Quikwriting always struck me as being closer to a keyboard in that respect:  until the user has used it enough to become expert, they are possibly going to spend more time looking for the letters than they will actually entering them. (The author also makes claims of it being a heads-up tech, which I'd also like to see results on, but I don't expect to, for no other reason than that heads-up tests just don't seem to happen, regardless of technology.  See, for example, the original Xerox paper on Unistrokes for another example of a technology that never had tests performed to see if it did, as the authors claimed, allow for heads-up input.)

Anyways, 6 years pass, and during a cursory search for new papers on stylus-based input, I came across a paper from NordiCHI 2004 which measured Quikwriting speeds on two platforms over a series of 20 sessions per user (for a total of 10 hours worth of training with the interface).  The results were, personally, not very surprising.

In short, it took them that many training sessions (although only half the time, or 5 hours, was spent using the stylus-based version—although part of the paper's premise was that the skills transfer between versions of the interface) for the entry-rate of the system to reach 16wpm, or on par with other stroke-based systems.*  (For contrast: the average handwriting rate in the same study was 26wpm)  (And for those thinking "I spend that much time on my Palm in a day": that's not 5 hours of casual use, that's 5 hours of nothing but text-entry.)

Ok, so it doesn't address whether or not, given enough training, the system is actually faster than Graffiti, and I'm sure that some people are more than willing to try.  However, as is, I'm more concerned with how easy it is for a new user to enter the text quickly, and with a minimum amount of learning involved, since, after all, users don't like having to learn new systems, especially if it's going to take a lot of time to become proficient.

(Hey, I finally finished writing the chapter of my thesis where I defend the reasons for choosing a unistroke-based system for input.  It should be expected that I'll slag all the other ones. ;-) )

If you don't understand anything I wrote, let me know and I'll try to elaborate.  Jo's always commenting that I use words and phrases as if I assume that the meaning is self-evident, when it may not actually be so.

* I say stroke-based systems as, interestingly enough, soft-keyboards actually have a very high theoretical maximum entry speed.  It goes like this:  once a user has had enough experience with the layout, the length of time they spend looking for the next key to tap is negligible.  This, therefore, can be considered to be so close to 0 as to be ignored, and the length of time to press a key becomes the length of time to move from one key to another.  Of course, this sort of thing has been very well modeled, so the length of time can be estimated.  The end result is that the theoretical maximum for a QWERTY layout is ~30wpm, with other layouts going as high as ~43wpm.  (All numbers taken from here.)

Of course, the "given enough time with the layout" is the same thing we have here: just how much time is needed before the users start approaching (or, in some cases, exceeding) these theoretical maximums?

Fitts' Law also has some interesting things to say about button size on soft-keyboards, but that's not really relevant to this discussion.