This is pure Colemak resemblance, and I feel I practically knew it already once I'd made it.

Yes, the $% and maybe '" positions do seem "too good" once you're used to thinking center in/out. But since the around-center drags work so well there are actually too many good positions on the pad! So in that context it makes sense to make the symbol mappings like $% consistent with our Colemak/QWERTY knowledge instead. That way the QWFPG/ZXCVB/JLUY_/DKM__ letter rows stand out well so they're easier to use for people who look at the screen while typing.

It's amazing to think of: With all tap/circle and drag/drag-return mappings on 9 keys minus the four corners and two shift positions, there are some 150 (18×9-12) possible mappings on the 3×3 ABC pad alone! And twice as many including the 123 pad. On a standard 102-key keyboard the character keys on the 4 main rows have 4 shift states each including AltGr mappings which gives us 192 (4×12×4) mapping positions but only half of those are in frequent use.

No unfortunately not.  Officially there's the Yahoo group, which was once very active, but it's dead now.

I asked Saied about a community forum but he pointed me to the Facebook group.  There's also @MessagEase on twitter.  But neither of these really accommodates broader discussion like a forum does.

Yep, still working with it and I don't think there are better alternatives for those who distrust AI. There are now a whole lot of people typing 60+ WPM with MessagEase and that's pretty impressive.

I think the 9 tap letters are actually very well placed to minimize travel distance (especially important with one-finger typing)! T E S need to be close to the Space because their letter-space bigrams happen all the time, TH HE ER AN NI are other very common bigrams (keeping in mind that they can go both ways so for instance ER represents both ER and RE) etc.; MessagEase's ANIHORTES layout puts it all together very nicely.

I've been thinking a lot about placing T in the middle because it figures prominently in a lot of bigrams and suffers a bit in the lower corner of ANIHORTES. The main reason for this is that T=H to the left seems suboptimal for two-finger typing. It would lead to much better connectivity to the other 8 letters and the 3 most common letters all in the middle column, but the big question is whether that'd quite make up for the longer travel between T and space! My best solution so far is ANI OTH RES, I think. Getting A to the bottom row would compensate for much of the loss of T there, but E needs to be close to both R and H while A=N=I is also a good idea so it's not easy to pull off.

| A | N | I |    | A | N | I |    | A | N | I |
| H | O | R |    | H | T | R |    | O | T | H |
| T | E | S |    | O | E | S |    | R | E | S |

Trying to get T in the middle: The T-O swap (middle) is the simplest move but loses O=R and O=S; ANIOTHRES (right) has many fast common bigrams

When it comes to the rest of the letters, I agree with you that they feel more haphazard and that's a bit of a pity. I don't think there's much speed to be gained from their placements (although some of them like D U etc should have decent places, most of the drag positions seem about on par with each other), so they should've been placed in a way that's easily learnt and remembered instead!

I do recommend using my 'Colemakoid' mappings of the secondary letters and symbols, as I find them a lot more intuitive to learn and remember. I type quite comfortably with these now, but ghen types a lot faster than me using the standard setup (probably not due to layout differences but to his awesomeness and better training - nobody knows for sure though).

Send me your email address in a forum mail if you want more info and off-forum details...

The ANIHORTES layout is supposed to minimize finger travel for English text.  The whitepaper describes how they calculated that (using Fitts' law), maybe you should try to reproduce those calculations as an objective measure for your alternative layouts?

For the secondary letters, I agree the standard layout is not very intuitive at first, but neither are Colemak or Qwerty. ;-)  But after a few weeks of practice you get to know it well enough.  As with touch typing, using the blind keyboard really helps.

Replacing the standard phone keyboard with a Colemak one shouldn't help, no. I never did that myself.

As you can see above there's also a QWERTYoid mapping which is quite similar to the Colemakoid one. And if you don't trust my claims as to how easy it is to use, then I'll just suggest you try it out yourself (if you're interested enough). Having worked with and thought so much about Colemak myself, I have both a strong visual and a motoric impression of the layout, but I do feel that the motoric impression benefits as well.

My "offer" wasn't one of discussion but of some implementation help that really doesn't belong here. But we'll just wait and see then.

How fast is 8pen? I'm now at 33 WPM with MessagEase, and Geert/ghen types a staggering 61 WPM with it!

Have you read this review? This guy has tried both and preferred ME for its versatility, ease and speed it seems. Interestingly, most of his gripes with ME have now been fixed. All in all, the searches I made when hunting for a touchscreen input method convinced me that ME was the most promising alternative currently available.

Since I recently changed my HW keyboard phone for a touch screen only, I took a look at MessagEase and compared the default layout to DreymaR's "Colemak" version. I started with DreymaR's, since I thought it'd be more familiar and easy to learn but after memorising the layout before the games, and one-two minutes in, I found that if I want to be efficient, I shouldn't try to make a link between Colemak's position of the letter and where should I be looking for it on the 3x3 keypad – it was just too slow. After maybe half an hour I figured all my speed comes from memorizing the motion for each letter and I felt no advantage whatsoever from knowing Colemak. On the other side – it was only awkward because I couldn't follow the learning method of the program (tap → centre drags → outer drags...).

So I tried the default layout and it felt quite nice. I also found an interesting logic of the arrangement which helped me memorise almost instantly the centre drags:
|q|u|p| → Q and P look the same, just mirrored
|c|o|b|
|g|d|j| → G and J sound the same and they're on the corners
    ↓
U and D remind me of Up and Down

Don't know how much of this was intentional but it does help. Deffinitely the best typing option for touch screen. And I love the nifty features – cut/copy/paste; select all; backspace/delete (+whole word); moving the cursor one step or a whole word – I'm constantly using all of these and it's awesome to have them so easy to reach.

My only complain is that some positions are not configurable and they seem to differ on the different layouts (I use English and Russian). It could be device-dependent, though, and they're all either the ones at the far top or far bottom.

On the computer I use Colemak for English and a custom layout for Bulgarian. On my old phone I used Qwerty since the benefits of Colemak don't apply when you're using only thumbs, and I have the Qwerty labels on the keys anyway.

It's been too little time with MessagEase and I've only trained English so far, so I can't tell you how well it'll go. I would try a phonetic layout but I can't change the base characters. I can say that I'm not looking forward to learning two layouts for my phone but that's the plan for now. What do you mean by Rulemak-like translit? Coud you explain a little bit more?

Oh yes, I see. I did use almost the same when I started with Colemak, but after I got used to Colemak, I wanted something optimised for Bulgarian as well. The official layout is optimised but follows bad rules so I designed my own, which I've been happily using for quite a while now.

But if you use it only occasionally, it doesn't make sense to learn a new layout, of course.

Given that I don't type that much on my phone, I'd have really loved to 've been able to make a "phonetic“ layout for cyrillic, but... I guess I'll do it this way.

Hm... I'll give it some thought when I get home. Reading this got me a bit interested now :)

Some more results, after running my program over all 362.880 permutations of the ANIHORTES layout: ANIHORTES isn't the best layout according to my metric, but it's in the top 0.7 percentile, so really good.  The winner is NOR IHE ATS (or its mirror image RON EHI STA), and has 4.8% less finger travel on my test corpus.  But the difference could just as well be attributed to the error margin of my oversimplistic metric?

The worst layout has "only" 29.2% more travel, which is still ok (compare that to Qwerty having 120% more finger travel than Colemak).  So the main optimization is in putting the most frequent letters anywhere on tap positions, and you'll end up with something decent.

So DreymaR, let's just stick with ANIHORTES. ;-)

Things got more interesting once I considered moving the space around as well, and thus put a letter on the space bar.  With this additional degree of freedom, and after 6.5 hours of number crunching (10! or over 3.6 million permutations), 20% of layouts score better than ANIHORTES.  The winner "HER T_S IAO [N]" has 15.4% less finger travel.  All of the top-700 layouts put the space either in the central O position, or the bottom row E position.  To my surprise, the layouts with space shuffled around don't get worse than the standard ones.  The worst positions are at the bottom (spacebar) and the top corner A and I positions.

Yes, my formula is surely oversimplified, but since I'm comparing just main letter layouts, I could afford to neglect secondary parameters like pressing time (which would be the same for all tap letters).  So even though my results aren't absolute in terms of wpm speed like ME's, they can be used to order layouts relatively to each other, which was my goal.

Also, now I confirmed (approximately) the same result using a different approach, which has value in itself as well.

Yes, dual finger typing will be much harder to model.  Contrary to touch typing on a physical keyboard, you can't assign a particular finger (L or R) to each motion.  And simply alternating fingers won't yield good results either.  The mind makes its own optimization for which finger to use for each letter, depending on the sequence it's in, and probably influenced by handedness and mere individual preference as well...

If it's really slow, then it should be a no-go. I have rediscovered the joy of typing on my phone; I hardly thought that could happen! No I no longer feel confined when I have something lengthy I want to write down. I've been sitting with my Galaxy Note jotting down various song lyrics, just because I could and it's fun (and yes, because I wanted to rehearse those songs of course). It's the Colemak-y joy all over. ;)

Also, I don't think 8pen is as flexible? MessagEase can actually be used on a number pad; it was designed with those in mind! Of course, then a 'tap' letter becomes a double-tap which isn't so efficient and could get cramping after a while, but it's still possible to type quite fast on according to the creators. MessagEase could be used well on lots of different surfaces, even fancy credit cards! And it's nearly as good whether you use one finger, two fingers or a pen. And you can enter lots and lots of different letters, accented letters and symbols. Ingenious, powerful and flexible.