Interesting thoughts. I suppose that when you say things like 'find a cool home row' you're making it a first seed and then allowing a little optimization by the algorithm? The hard part will, as always, be to determine proper weights and priorities I suppose.

Just to illustrate how tricky it can get, consider alternative fingering! Since same-finger typing is so bad, good typists will often circumvent it with tricks. I'll type 'uhm' with middle-index-middle fingers (in Colemak, of course!) thereby saving effort.

My most used alternative fingering is 'nk' and I'm quite safe using that. On the other hand, I do slow down on alternative fingerings that I'm not using a lot. As I said, it's tricky. If I were to type 'uhm' a lot, the problem with it would matter more but I'd also be better prepared to cope with it.

That's one major reason why QWERTY still rules supreme I think: Good typists can get around some of the worst flaws, so they don't feel the need for a better layout as intensely as we layout freaks tend to think they ought to!

Thing is though, I know that Arensito and Dvorak and Colemak are good layouts and they were all made considering some design principles that I consider sound. Thus, putting that into a machine will make layouts that are almost but not quite 'Dvolemaksito', won't it? I don't really foresee that it'd surprise me in a positive way; I'd create a slew of layouts that were 'almost-as-good-as-and-maybe-unnoticeably-better-than' the original favorites I started out with.

Maybe I'm just lacking vision, heh. But I'm doubting that this method will innovate enough, and furthermore it'll be hard to whip up objective and controllable success criteria for it (but that's true for any method of course).

I think this point nails the problem on the head.

The difficulty isn't solving the problem, it is posing the problem. We've all, myself included (http://mkweb.bcgsc.ca/carpalx) made attempts to solve it by posing it a different way. Therefore, I think we should not be arguing layouts as much as the criteria that goes into a layout.

Given that beyond the basics, subtle qualities of a good layout are actively disputed, any standardized test that evaluates a layout is only as good as its assumptions about what is good.

I, for one, do not weigh aesthetics into a layout, since a layout should be easier to type on and not easier to look at. A standard QWERTY letter mask may attract more adopters, and in particular if some of the edge or home row keys are the same, but overall cosmetics simply makes a layout more familiar rather than less strenuous.

The speed at which the algorithm runs should not be a factor in determining its usefulness. CPU cycles are cheap and once a solution has been found, we need not look for another one. Of course, that is, untill the next layout algorithm is developed :)

I agree: One of the strengths of Colemak is that it appeals both to the enthusiasts and the laypeople (well, the few laypeople who give a damn about layouts at any rate...). It isn't just a layout for the select few, potentially.

The punctuation placement in Dvorak put me off a bit, and designs that let punctuation fly everywhere even more so. The exception I will make is the key between B and K - if you have such a key (Comfort layout). That can be practically anything since it shouldn't be hit much and is clearly placed between the two hand blocks.

Yep, I'm loving my funny-vowel key!  :D

Since I'm using PKL for the most part, my space bar doubles as my second Enter key - and it's a lot easier to hit than that tucked-away former B key! One of my reasons for demoting that key to funny vowel usage was its difficult position so it shouldn't be used for anything more common than, say, the letter 'J' (which in the other somewhat tricky position between the hands).

You really find Colemak-semicolon easier to strike than Colemak-K?  The former involves moving the shorter and weaker pinky, considerably tougher than swinging the longer and more powerful index finger a little bit down, no? 

Somebody should thrash out statistically the relative ease-of-strike values for all the key positions.  I am not sure the implicit assumptions behind Colemak and the explicit ones made by Martin K are beyond revising.  Colemak design, for instance, seems to assume that Colemak-D is easier than Colemak-F, and Colemak-G is easier than Colemak-V, which I cannot reconcile with my experience.  Colemak-G is quite misplaced in my opinion, and could have been left at its Qwerty position, strengthening the layout's claim to be reassuringly similar to Qwerty.

That said, I do realize that if does make sense to occasionally gift an easier key to a less common letter, in order to not to work the strong fingers too much.   But the G and D positions were sufficiently achy that I reluctantly had to switch some five keys around for personal use.

Using Dvorak, Carpalx (QGMLWB), Klausler, Capewell, Colemak, and MTGAP 2.0, I calculated the average costs of different positions.

144 105  81  77 103 131  78  69  82 121 
 22  20  19  29  63  61  27   7  31  30 
165 153 103 122 155 131  97 105 125 154

And if you find those large numbers to be intimidating, here is a normalized version:

 23  16  11  11  17  27  11  11  13  25 
  2   3   6   2  14  13   4   1   5   3 
 27  26  13  19  24  21  15  17  21  24

This is biased, of course, since each of these layouts takes other factors into account when placing keys. But it's good overall.q

The way I did it was I made costs based on what keys were placed where in the layouts that I mentioned. e is +0, t is +1, a is +2, o is +3 . . . z is +25. Since so many keyboards place e under the right middle finger, then yes, the left middle finger is supposedly six times harder than the right one. But you should really measure it in absolute cost instead of proportional cost.

Let's say Colemak places e under the right middle finger (which it does). This position gets +0, since the designer of Colemak must think that that position is pretty easy. But Colemak places z under the left pinky on the bottom row. This position gets +25, since the designer of Colemak must think that that position is hard to hit. Repeat this process for every key on the keyboard and for several different layouts, and you get a pretty good average of the costs of different positions.

No, but it's still useful as an estimation technique.

I tried doing that about a year ago, but it didn't get results that were anything like what I wanted. My implementation was probably not very good. I find numeric values to be efficient and easily readable by the computer, and (with a little calibration) can get very good results.