I definitely agree a language model is key to better input schemes beyond the current state of the art on mobiles .For instance RIM/TrueType, iPhone, T9 seem to only utilize a dictionary based solution as opposed to using a full fledged language model limiting their effectiveness.
That said, while I like the language model I just don't se this interface being extremely usable. I believe there is a new interface waiting to be invented that, with this language model technique, start to approach keyboard typing speeds. Swype seemed on the right track, especially by reducing the learning curve and leveraging a QWERTY model, but I hope that even better options develop. For now a slide-out keyboard on my iPhone would be good enough for me. The PalmPre may even convert me if it works as good as it looks.
I think the purpose of Dasher and the other input techniques mentioned is to be able to quickly enter text in constrained situations, whether its due to size restrictions of the device or due to limitations on the input vehicle (e.g. single finger, one handed, thumbs only, etc). I think in any of these situations, I would be more then happy to be able to enter input as fast as I could type on a standard keyboard.
Now if someone could invent a faster then keyboard text entry system that had a decent learning curve, I would be all in. I tried Dvorak and never was able to type faster then on Qwerty, plus it took way too long to adapt too
That aside, I've heard of dasher before and tried it out on a few occasions. I like the idea, but wasn't impressed enough to use it for more than a few trials. It might have been different if it was a fully integrated keyboard on the PDA rather than a separate app.
One thing about using it is that it's not fast. There is (was) no way to autocomplete a word as the iphone does - you're stuck either trundling or doing a speed-up-slow-down-whoa-whatnow dance which isn't helped by the lack of clarity of what's coming next. There are lots of letters flying about. You're faced with doors a-z and every time you go through there's another set of doors a-z, each one in a different position, size and colour to last time.
There actually is something like auto-complete. If something is a common word or syllable, it is possible to zoom through a series in rapid succession. The whole thing is like a huge auto-complete.
This would be awesome with a really well engineered device like the IBM pointerstick. With a mouse it's so-so.
Got to say, this thing has potential. I was typing about 12 words a minute after tying it for a short while. Kinda torn about the predictive nature of it, though. On the one hand, it lets you type faster in the beginning, but on the other hand it makes it harder for you to really speed up, as it is akin to having the 'g' button go huge on your keyboard right after typing the letters 'havin' but still. I was typing at 5 wpm when I first got onto a computer (I was 4) and now I type at around 40 to 60, after thousands of hours of experience.
If you are going to give it a try, definitely ramp up the speed or you are going to judge it on false pretenses.
if you want to read a good mathematical exposition, check out chapter 6 in Mackay's (the guy giving the presentation) book "Information Theory, Inference, and Learning Algorithms". It covers stream codes and, specifically, arithmetic coding.
I've recommended this book in other threads... its an amazing read.
That said, while I like the language model I just don't se this interface being extremely usable. I believe there is a new interface waiting to be invented that, with this language model technique, start to approach keyboard typing speeds. Swype seemed on the right track, especially by reducing the learning curve and leveraging a QWERTY model, but I hope that even better options develop. For now a slide-out keyboard on my iPhone would be good enough for me. The PalmPre may even convert me if it works as good as it looks.