Well I remember the paper or article (damn shame I forgot to bookmark the thing) suggesting for some reason that the brain can easily pick this up and adapt.  And if that's the case then why not run with it?  Problem is that it only works for veteran touch typists that have it already wired into their brains.

My suggestion was to just use more predictive text.  Seeing as Dvorak pretty much puts the vowels on the left hand, you could probably just omit the vowels and use the right hand only and let software guess the words from patterns.   The left hand could then be freed up for something like selections, modifers, numbers and directions.

I never imagined that my typing would get as fast or faster than auto-completion.  By the time I wait for a suggestion to popup I may aswell have typed it.  Suggestion systems can't afford a lag and need to be very fast to compete with fast typists.

My laptop has enough ram to hold an entire dictionary, so I don't really see the problem here.  As Dasher is open source, it's possible to pick out some of its components, like the predictive text stuff, and for you to roll your own.  Speed shouldn't really be an issue on modern hardware.

I'd imagine that if you were presented with a prediction as you type, space bar would be a good way to move along to the next word (picking from alternatives would probably need an easy selector).

The built in spell checker for vim is good but it's really awkward to use because of the key combos (I could remap them...)  as such it's pretty useless to me.

I've a terrible habit of not proofing what I've read.   Perhaps we should just concentrate on writing shorthand, and then flowing through the text afterwards transforming the shorthand into real words, and doing our corrections while we are there.  The post analysis of what you write can be the lengthy bit anyway - so perhaps we shouldn't get too hung up on trying to get it right first time.  We are not all courtroom transcribers, and those that are, are probably using steno!

Other forum members  recently championed  Pentatdactyl (whatever it's called) for Firefox and Vim for desktop nirvana.   But at the end of the day,  they are only two applications, what happens when you open another app that requires text input?

Ideally you want a consistent input interface for text entry across all your applications at the Desktop/OS level.  Android from my brief flirtation with it, kind of does this with it's text entry system, you can toss one out for the other.  But I don't see anything like this for the desktop.   For me the OS should be far more clever about this.

The text input method needs high priority, and most likely context support.  That might be language or task based.  Auto-completion may have different suggestion mechanisms in one application than another.

The interesting part about the idea is the ambidextrousness that our brains possess.  This knowledge may help in layout development.  A modifer that is mirrored might lead to easy ambidextrous use (Shift, CTRL etc).

@pinkyache

You misunderstand. It isn't that the autocomplete is slow, it is that the human reaction time is slow. For a competent touch typist it is usually faster to complete the words by hand from muscle memory than it is to break your train of thought responding to autocomplete prompts.

Here, take this example. I want to type 'three dog'. So I put down a T, and the auto complete list pops up with some options. I do not stop typing while reading these options (why would I?), I am still putting down letters. Three probably doesn't even show up on the list until I am up to R at which point my finger is already on the downstroke for E, my other fingers are half way to where they need to be for the next word (index finger sliding under the middle to rest above D), and I am deciding on how to word the next sentence.

For slow typists, also a bad match. Most hunt and peckers (that is what we mean by slow typists, right?) don't look at the screen while typing, so any options provided by an autocomplete would be lost on them.

There are exceptions. Legal, medical or other technical settings, with big scary words like, I dunno, antidisestablishmentarianism. Yeah, sure. You can autocomplete the hell out of that. But do you really want professionals using autocorrect or autocomplete to fill out your medical files?

Second exception is if you are using something that autocompletes whole phrases instead of words. I find these to be of limited use.

Also, space is a poor choice of selector. You type 'the', auto correct suggests 'theater' and before you know it, 'the show tonight' has become 'theater showtunes tonight'. LetMeType that I mentioned earlier uses number keys, which aside from the fact that reaching up for the number row slow and clumsy, works fairly well. Outside of l33t speak and passwords, numbers are not usually used in words.

Cevgar:

> You misunderstand. It isn't that the autocomplete is slow...

(It certainly can be laggy).   I believe you could get used to a different flow.  It really depends on a good implementation.

> For slow typists, also a bad match. Most hunt and peckers (that is what we mean by slow typists, right?) don't look at the screen while typing, so any options provided by an autocomplete would be lost on them.

Hunt and peckers half look at the screen (they snatch glances).   Auto-complete can certainly help them.

> ...LetMeType that I mentioned earlier uses number keys..

Not my first choice. Perhaps doable on Dvorak given the left hand has less to do. ( Have to give this some serious thought at some point...)