One issue I've always had with QWERTY (and probably this new layout as well) is that "similar" letters are often next to each other on the keyboard, making typo detection nearly impossible in English. For instance, in English, many short words with a "i" as the only vowel remain a valid word when the "i" is replaced by an "o" or a "u" -- pit/pot/put, bit/bot/but, tip/top/tup, pig/pog/pug, bitch/butch/botch, etc. This means if you go to hit the "i" key but accidentally hit the "u" or "o" keys, the typo won't necessarily look like a typo to either to the author or the reader.<p>Similarly with "s" and "d" -- pretty much every verb can be conjugated to end with either an "s" or a "d", so if you press "d" when you meant to press "s" or vice versa, it probably remains a valid verb, and thus is harder to catch.<p>For these purposes, my ideal layout would have interchangeable letters sufficiently far apart from each other to make typo detection+resolution easier. The most commonly used letters should be surrounded by the least commonly used letters so that, for instance, if you "jqb" I know you meant "job" and not, say, "jib".<p>This would also make swipe typing on phones far easier to use, as it would be much less ambiguous which letters you were aiming for.
Y'all need to know the background of past efforts [1], people keep making the same observations over and over again as if they're new. In the area of computer-generated layouts, there's been lots of rich discussion on the AdNW [2] and MTGAP [3] projects.<p>For "improving Dvorak", that was solved back in 2004 with Capewell-Dvorak [4].<p>To the author: what is supposed to be AI here? I don't see a mention of a ML method or neural network here, is it "just" an evolutionary algorithm like CarPalx, AdNW and MTGAP (not that it's bad)?<p>[1] <a href="https://deskthority.net/wiki/Alternative_keyboard_layouts" rel="nofollow">https://deskthority.net/wiki/Alternative_keyboard_layouts</a><p>[2] <a href="https://deskthority.net/wiki/Alternative_keyboard_layouts#AdNW_.28Aus_der_Neo-Welt.29" rel="nofollow">https://deskthority.net/wiki/Alternative_keyboard_layouts#Ad...</a><p>[3] <a href="https://mathematicalmulticore.wordpress.com/category/keyboards/" rel="nofollow">https://mathematicalmulticore.wordpress.com/category/keyboar...</a><p>[4] <a href="https://deskthority.net/wiki/Alternative_keyboard_layouts#Capewell-Dvorak" rel="nofollow">https://deskthority.net/wiki/Alternative_keyboard_layouts#Ca...</a>
The problem with any non-QWERTY keyboard layout is that all app and web shortcuts go out the window. You want to save an image for the web? Good luck growing a third hand. Want to open the quick menu on a site using ⌘+/, but your / key is only accessible using the shift key? Tough luck. Need to change the font size in your editor? Just navigate down these 3 easy sub-menus, because the keyboard shortcut for that is already reassigned to alleviate one of the other 300 broken shortcuts.
It occurs to me that optimal keyboard layouts depend primarily on use cases: by human language, by activity, and so on. I might well imagine that a Perl-optimized keyboard layout would look different than one intended for writing fiction, and so on.
As we are discussing layouts, I would like to point to a bigger problem (in my opinion) with keyboards: they don't have enough keys.<p>Let me explain. While Latin alphabet has only 26 letters, many languages have more of them. For example, Russian alphabet has 33 letters. Therefore, to fit these letters there must be 7 more keys. But we also need two keys to switch layouts - so in total there should be 9 extra keys.<p>Sadly, manufacturers (including Apple that claims to "think different", but in this case thinks exactly the same as PC vendors) don't want to add even a single key, instead they put extra letters onto keys with symbols, so you have to switch layouts to type those symbols. Furthermore, several characters like period or comma are located on different keys in different layouts. Guess, how easy it becomes to make a mistake.<p>Switching layouts is inconvenient. Instead of having two dedicated keys for switching to each language, OS vendors add single switching combination like Shift + Alt (Windows) or Cmd + Space (Mac). Also, on keyboard itself there is no indication that these keys can switch layout.<p>One could argue that it is difficult to find more space on laptops. Well, I think that this can be solved somehow. For example, one could add an extra row. But it is absolutely possible on desktop PC keyboards, and yet nobody is doing this.
This feels like over fitting if it's not allowed to make basic changes like a capslock being control when held and escape when used on its own.<p>There's a bunch of things like that which would have a bigger impact if they could be measured but are ignored by this methodology.<p>I use standard Dvorak, but thumb keys are the real game changer (the Kinesis introduced me to them) so CMD keys that act as backspace and enter when used alone are good, as is using space as a shift key to get numbers and symbols.<p>Two keys (j+k) at once for esc and : and home row mods are other quantum leaps.<p>Oh and the data is from non-ortholinear keyboards, so will be skewed for non staggered keyboards.
Alternative keyboard layouts are fascinating, but always raise a couple of questions for me and I wonder if anyone here can come up with actual experiences regarding the following topics:<p>- I have been using standard german QWERTZ for about 42 years now, in average certainly not less than 3 hours every day. I was originally educated in proper 10 finger blind typing in a dedicated typing school as part of my typesetter apprenticeship. I am typing really fast. Today I am even using "daskeyboard ultimate" - a completely unlabeled keyboard. Is it really possible to improve 42 years typing practice with a new layout?<p>- Is it possible to be proficient in two keyboard layouts at the same time? I have a partial answer myself. For one, I managed to learn standard Emacs shortcuts (a lot of them!) some of which contradict common OS shortcuts (like copy/paste etc) and I have both sets in muscle memory and I can seamlessly switch between both sets as I switch applications (most of the time that is). And I use three computers with three OSs (Linux, Win, Mac) with one keyboard per Synergy software. Same layout for the standard characters, but different OS specific layouts for special characters like "@", "€" etc. I can seamlessly switch between them as well and that is the reason for the unlabeled keyboard. A labeled keyboard would show only one layout and puzzle me if I had to enter special characters for another. So, while I know I can handle some number of characters in alternate layouts in muscle memory - will this be possible for the entire character set as well?<p>- If you get accustomed to a non-standard keyboard layout, isn't it a problem if you have to use some other computer (where you cannot change the layout) for whatever reason, for example in an Internet cafe?
Apart from the technical merits, I always appreciate "hacker" naming of software.<p>"The name is a combination of HAL-9000, as a reference to the layout being designed by an AI. And, Dvorak as a gratitude to Mr. Dvorak for his dedication to the layouts optimizations process. The letter m in between is just to make it sound nicer. Or is it!?..."<p>I suppose the "m" is for MadRabbit, the author. Nice touch.
CarpalX [1] is also a keyboard modeling approach that includes typing effort and corpus loading. I would love to hear about the difference in typing effort modeling and approach for CarpalX.<p>CarpalX is a linear parametrized model with configurable parameters and weights for different types of strokes, so you can weigh the effort. You on the otherhand have made measurements of your hand movement. Very impressive and valuable! In the explainer [2] you mention the datasets briefly skipping over specifics:<p>- Are the hand-movement metrics a measurement of the center of the hand from an video-based object detector?<p>- Do you have individual finger detection too (doubt it is very precise due to occlusion)? My guess is relative hand movement is a good approximation of overall finger movement though.<p>- Any typing effort model based on this data will not be applicable to ortholinear or alternative layout boards such as a split hand (Ergodox, Corne, Moonlander or Plank).<p>- You briefly mention removing any manually determined effort-based objectives (like those in CarpalX) from the genetic algorithm optimization. You say they are highly similar to the purely trigram data-driven approach, but I am still very curious to see the results with these objectives included.<p>I am not trying to put you down though: It is still very commendable to research a better layout for the most common staggered 60% lay-out.
I have been keylogging myself for three years now and will definitely try out your approach though, looks very promising!<p>1. <a href="http://mkweb.bcgsc.ca/carpalx/" rel="nofollow">http://mkweb.bcgsc.ca/carpalx/</a><p>2. <a href="https://www.youtube.com/watch?v=pZ40gmfDFfQ" rel="nofollow">https://www.youtube.com/watch?v=pZ40gmfDFfQ</a>
An interesting related project is the AdNW (<a href="http://www.adnw.de/" rel="nofollow">http://www.adnw.de/</a>) layout, which was created by defining metrics based on Dvorak’s ideas and running a genetic optimization algorithm on a corpus of German (50%) and English (50%) text.<p>The symbol layers were copied from the Neo2 layout, which afaik were arranged manually.
AI decides that ctrl is still there? Look at many Japanese made (ANSI) keyboards: they simply put ctrl where caps lock locates. This is a life changer to me, and inspires me to change all my keyboards: no new layout required.<p>Even though they can claim movement is decreased, having little finger hitting ctrl at the original location for hundreds to thousands of times is never a fun. It’s ergonomically bad.
Not as good as this!
<a href="https://www.charachorder.com/" rel="nofollow">https://www.charachorder.com/</a><p>500wpm apparently!
Dvorak touchtyper since 2010 is here ^_^ I appreciate any attempts to make August Dvorak's work even better. For example, that claster of wovels is not AI's work but Dvorak's. BTW there is one strange claim from my POW:<p>> QWERTY - 0% (baseline)<p>> Dvorak - +77%<p>> Colemak - +84%<p>> Workman - +101%<p>> Halmak - +134%<p>If author considers Colemak and Workman more efficient than Dvorak, I should not even to give this layout a try :) because I have tried these and can not report such experience. Probably it is because I am a piano player since 2016 and the fingers which are weak on average people are not that weak for me. Another rebuke to mention is that Halmak is not compatible with Programmer Dvorak. That idea of 7531902468 and three pairs of brackets, braces, parentheses (all paired symbols except diamond braces) is extremely handy while programming.<p>AFAIK, 7531902468 is also August's gem of mind, can your AI/HAL/whatever to design a layout with that row and at least [{}(=*)+] cluster untouched?
After using a split spacebar for some time now, I just can not see the idea of a single spacebar being optimal.<p>I have two thumbs. Why not use them? My right spacebar is space, my left is enter. I have never done anything else so profoundly productive to my computer.
fwiw, running the same evolutionary algorithm on Julia, Python and C++ files returned this layout<p><pre><code> ` 1 2 3 4 5 6 7 8 9 0 - =
c l r b z ; q u d j [ ] \
s h n t , . a e o i ' \n
f m v w ? y g x k p
11 7 13 14 13 13 9 19 , 45 | 55
Symmetry: 61 Evenness: 81
Overheads: F:37%, H:12%, S:9%</code></pre>
After having tried Dvorak for a couple of days (weeks?) a couple years back, I found my peace with the american default layout. Typing on dvorak was fun, and the learning process was really fun, too - fighting two decades of muscle memory was quite the experience.<p>But at some point I realized that I now have to adapt all the bindings - without a solution to this, I'm not willing to try out different stuff. I need vi to work out of the box, especially because I also use the vi-bindings in tmux, zsh and firefox. Sigh.<p>Looks really interesting though.
Dvorak is one of the well-known layouts and even now we can't have it on the iPad, so I don't think keyboard layout is only about efficiency it's also popularity and availability.
I used the QGMLWY layout by Carpalx[0] for a year or so. The site is really interesting, worth a read. Afaik they made a list of the most common trigrams (three letter combinations) then used a genetic algorithm to optimize the layout for most of the same factors listed in OP's GitHub Readme (minimizing same finger sequences, certain kinds of movement).<p>In the end I switched back to qwerty for 3 reasons:<p>1. keyboard shortcuts ending up in weird places and behaving inconsistently between programs (at least on Windows, some of them seemed to respond to the physical key rather than the letter).<p>1b. so for games i'd be used to WASD but WASD ended up in different places so I'd either have to remap the keys in every game, or switch to QWERTY while
using them. Same with browsers, the Ctrl+N and Ctrl+T ended up in weird places I somehow never got used to.<p>2. different preferences in terms of how my hands work and how comfortable each key is to access. (e.g. i'd rather move up a row than reach to the middle of the keyboard. Minor but it adds up with use.)<p>3. frequently switching devices. I think there was no QGMLWY support for MacOS (but I also recall rearranging my keyboard keys -- perhaps I had switched to Colemak at that point because it was built in.)<p>[0] <a href="http://mkweb.bcgsc.ca/carpalx/" rel="nofollow">http://mkweb.bcgsc.ca/carpalx/</a><p>Also somewhat related, regarding programming comfort specifically, something I thought was cool is swapping the numbers and symbols in Vim, so you can access symbols without Shift. I don't use Vim but I kind of wish there was a system-wide toggle for that kind of thing. (Maybe with AutoHotKey, and then porting it to HammerSpoon or whatever the cross-platform one was called :)
Has anyone already learned Dvorak and made the switch to some other "more efficient" layout, such as Colemak, Workman, Neo, this new one, etc? Has it actually been more ergonomic, efficient, and/or faster than Dvorak?<p>I have a Moonlander and have it totally tricked out so I have no problem with qwerty hotkey locations. It's basically just the time delta to grind a new layout. When I learned Dvorak around 2009, it took me about a month to reach parity. It might be even faster this time around since I'm constantly messing with my layout with small tweaks and having to relearn muscle memory.<p>Also if you aren't using some kind of layer scheme to map things like ~./()= up/down/left/right to home keys, you are really missing out :p
Interesting to compare it to my preferred keyboard layout, BÉPO [0][1].<p>In Halmak, the vowels are clustered around the right hand rather than the left. Like BÉPO, the punctuation is reachable with index fingers. There is some difference with the home row consonants, but otherwise this looks like an interesting layout.<p>Although BÉPO is a French layout, I’ve found it works well for English – despite giving space to accented characters that are not often used in English.<p>[0] <a href="https://bepo.fr/wiki/Accueil" rel="nofollow">https://bepo.fr/wiki/Accueil</a> (in French)
[1] <a href="https://en.wikipedia.org/wiki/B%C3%89PO" rel="nofollow">https://en.wikipedia.org/wiki/B%C3%89PO</a>
I'm three months into colemak now, together with learning touch typing (and switching to vim as main editor).<p>I've managed to get up to around 50 wpm coming from 75 wpm with normal qwerty without touch typing (just 6 finger freestyle). I've still some time to go to get over my initial typing speed. A really cool thing about colemak for me is replacing caps lock with backspace, which in itself is getting rid of so much finger/hand-travel, but that can obviously also easily be hacked into qwerty as well.<p>To this point, I don't know if I would recommend switching to anyone for the following reasons:<p>- You will slowly use muscle memory of normal qwerty, which can become quite awkward whenever you forced to use another computer. It's not like you cannot type it anymore, but you will be quite slow and inaccurate when typing. However, these situation barely exists for me in everyday life.<p>- Learning the new layout is quite a feat that takes time and daily dedication. I decided early on to use the new layout in my job (frontend dev) which definitely speed up adoption for me but also slowed me down considerably for some days. Even then it will take quite some time to get back to your initial speed.<p>- Depending on your profession, typing speed may not at all be a bottle neck. This is true for me as a software developer, where you spend the most time thinking about how to solve problem before typing them in small chunks.<p>- Wrt touch-typing, I weirdly found out for myself that it can actually cause some wrist and hand strain rather than protect from it. To me it feels that by using a lot more muscles to type it also increases chance of wear-and-tear. This is especially true for the pinkies for me, which I never used much for typing before.<p>Good thing about colemak wrt keyboard shortcuts is that it only changes letters (no symbols, punctuation, etc.) and then as few letters as possible to still achieve the best finger travel. In practice that means that many shortcuts stay the same, e.g. the common ones as CTRL-Z/X/C/A/Q/W etc.<p>I sort of did the switch as a self-experiment after being intrigued by all the science behind optimized layouts, and also to challenge myself to learn a new skill for the new year. I wanted to learn touch typing after decades of freestyle typing which I increasingly noticed was very error-prone. I figured that learning a new keyboard layout at the same time is a very good opportunity.<p>Not sure if I will stick to colemak permanently but so far it's still fun to try to gradually improve on it.
I used dvorak for two years got no benefit from it. It was no more comfortable than touch typing on qwerty, words per minute (a pathetic 80) and accuracy was the same.<p>Switching between dvorak and a qwerty keyboard was impossible even tho I knew how to touch type on both.<p>The big win from my dvorak experiment was learning that when coding typing speed doesn't matter. I spend more time thinking about my code than implementing it, removing shortened non descriptive variables or functions with long winded names has improved my productivity immensely.
people are saying it takes a year to reach proficiency with a new layout. tbh that seems like a really long time for a for someone who types for a living. + 10-15 wpm isnt worth it for me
Something I was developing on my spare time :) : <a href="https://github.com/aj0y/fitjoy" rel="nofollow">https://github.com/aj0y/fitjoy</a><p>The main difference with other layouts is that I take finger length into account. (Eg. the pinky is usually much shorter than your ring finger, so putting them all on the same row means your ring finger need to curl up more and less relaxed).
Also tried to reduce changes to the most common shortcuts we (programmers) use.<p>Feedback welcome!
When talking about keyboard layout, I'd like to see more experimentation in the physical layout beyond the 4 rows with 13 staggered columns, etc etc<p>I understand "keyboard layout" often means just changing the position of the keys on existing hardware, but perhaps we shouldn't be so limited, particularly when we can experiment with alternative layouts on touch devices.
As a chronically fat-fingered typist working in English, all those vowels next to each other gives me the willies - far too easy to get the wrong one and for autocorrect to not spot the error. The adjacency of U, I and O on qwerty gives me enough aggro in this regard already.
i rarely find myself in a situation where my typing speed on a regular layout is the limiting factor for input speed. When coding I don't type a lot and i'd likely get the same input speed on any reasonable layout after a while and when typing normal text i'm much more bound by how fast i can think about what i want to write. The only time i feel limited is when trying to transcribe while someone is talking but that's also a solved case since Shorthand and special typing equipment for that exists. I doubt you can type on any ascii based layout as fast as someone can on a shorthand typing system with equal training when it comes to natural language transcription.
Wonder how this compares to QGMLWB, which is also designed by a program.<p><a href="http://mkweb.bcgsc.ca/carpalx/?full_optimization" rel="nofollow">http://mkweb.bcgsc.ca/carpalx/?full_optimization</a>
I like it just because it will annoy fellow vim users who brag about never having to leave the home row to move around a buffer.<p>(You can remap, but you must also remap your muscle memory)