• You are not logged in.

    New "Dvorty-style" Layout

    • Started by Phynnboi
    • 10 Replies:
    • Reputation: 0
    • Registered: 17-Dec-2008
    • Posts: 59

    Let me preface this by saying that I hated Dvorak and have hated every other Dvorak-like layout I've tried.  This probably means that Dvorak fans will hate this layout, because I really like it.  But anyway:

    kldy g   q fbp;
    rnht s   u eoai
    zxcv j   w m,./

    Pros:
    * Row jumping (same-hand top-to-bottom and bottom-to-top) is very low (even lower than Colemak).
    * Persistent same-finger (e.g., "UFJ" on Qwerty) is okay (better than Colemak).
    * Finger load is well-balanced (each index finger does approximately the same amount of work, each middle finger, etc.).
    * Much easier on the right pinky finger than Dvorak.  Actually, is easier on the pinkies in general than either Colemak or Dvorak.
    * Preserves most of the Qwerty bottom row (ZXCV and M,./).
    * Actually scores well on that DDvorak test, unlike most of my layouts--is competitive with Colemak and Dvorak there.  (In other words, it survives a "second opinion.")  :)
    * Feels really good to this picky typist.  :)

    Cons:
    * Same-hand same-finger is rather high.
    * Load on the index fingers is heavier than I would prefer compared to the middle fingers.  The index fingers are together responsible for around 43% of all typing; the middle fingers, 22%.
    * Moves more keys than Colemak.

    I call it the Johnson layout because none of the rows spell out anything particularly compelling (and "Johnsok" just sounds--bad).

    I've been slowly learning the layout.  I'll report back when I hit 70 WPM with it, which seems to be the speed at which I feel "comfortable" with a layout.  It actually feels like I should be able to beat my old Qwerty speed (~105 WPM) to a significant degree with it, but we'll see.  So far, my main complaints are the positions of the "S" and "R" characters.  Unfortunately, there's nothing for it--I tried my darnedest to fix those (and the same-finger), but all the reasonable "repairs" made the layout feel significantly worse (often because they moved "H", which is pretty much in the perfect spot:  I believe "TH" is the most common digraph in English, and "THE" the most common trigraph; both are exceptionally easy to type on Johnson).

    Offline
    • 0
    • Reputation: 0
    • Registered: 08-Mar-2008
    • Posts: 303

    Interesting layout. I notice that the same finger is really bad in some cases; for example, ST, a very common digraph, is on the index finger. To your credit, most of the same finger is on the relatively flexible index fingers.

    I see the excessive same finger and the index overload as the major problems with this layout.

    I wouldn't trust the DDvorak site, because the algorithm is not very good.

    Results from running it through my algorithm:  (rightmost is best)

    Total keys pressed: 13332362

    Overall Score: Johnson = 20.815 < Dvorak = 20.086 < Colemak = 16.492
    Distance: Johnson = 90.. meters < Colemak 93.. < Dvorak 100..
    Same finger: Johnson = 3.01% > Dvorak = 1.2% > Colemak 0.6%
    Same hand:  Colemak = 22% > Johnson = 19.38% > Dvorak = 15%
    Same hand reverse direction: Colemak = 2.1% > Johnson = 0.6% > Dvorak = 0.59%
    Jumping over the home row: Johnson = 0.32% > Dvorak = 0.3% > Colemak = 0.28%

    (these are not compared, because they are less important and I am lazy)

    Reversing direction on 1 hand: 0.6%
    Changing rows: 10.37%
    Reaching to center column: 2.74%
    Inward rolls: 3.46%
    Outward rolls: 3.17%


    REVIEW: The layout has good distance, but sacrifices a great amount of same finger (with 3x that of Dvorak). The same hand reverse direction, which I consider more important than same hand, is very good. Home row jumping is not as good as Dvorak or QWERTY, but is close. Overall, this layout is strong in some areas, but IMO it is not worth it.

    Offline
    • 0
    • Reputation: 0
    • Registered: 17-Dec-2008
    • Posts: 59

    There doesn't seem to be a good way to eliminate the same-finger issue with a Dvorak-style layout without sacrificing finger balance and/or finger preference (i.e., index > middle > ring > pinkie).  For instance, both Dvorak and Colemak tend to use one of the pinkies more than one of the ring fingers (ISTR Dvorak even uses one of the pinkies more than one of the middle fingers).  Surprisingly (to me), even unlocking ZXCV,./; didn't seem to make much difference.

    I was able to get same-finger to near-Dvorak levels, but the resultant layouts either felt significantly more awkward or overworked one of the pinkies.

    I'm still not very fast with the layout, but I haven't found enough problems with it to give up yet.  (Not just because it's mine, either; I've given up on two of my former layouts already.)  Honestly, I hardly notice the same finger (except with "QUE", which is regrettable).  The heavy index finger use isn't noticeable, either.

    I really wish I could have captured into some heuristic what about this layout feels good to me, so I could maybe have made one with more pleasing stats, but I wasn't able to.  :/

    EDIT: I wonder how much it has to do with distance.  How do you calculate distance?  Maybe I'll add that to my algorithm and see what it comes up with.  Have you tried optimizing only for distance?  I'm curious what the lower bound is for your corpus.

    Last edited by Phynnboi (10-Jan-2009 14:48:38)
    Offline
    • 0
    • Reputation: 211
    • From: Viken, Norway
    • Registered: 13-Dec-2006
    • Posts: 5,345
    Phynnboi said:

    Surprisingly (to me), even unlocking ZXCV,./; didn't seem to make much difference.

    With the possible exception of the 'V', I can't say that I'm surprised at all.  :)  Nor would most of the others here be I guess.

    That had me thinking about one thing, which is that shift I keep harping on about (ZXCVB goes one key to the left, and the VK_102 to the old B position). It does reduce distance a lot for 'B', obviously - but the others not so much if at all. However, the effect on comfort is radical in my opinion. Only a little for 'V' but a lot for 'CX' and noticeably for 'Z' by my intuitive judgement. Would any of your scoring systems pick up that?

    *** Learn Colemak in 2–5 steps with Tarmak! ***
    *** Check out my Big Bag of Keyboard Tricks for Win/Linux/TMK... ***

    Offline
    • 0
    • Reputation: 0
    • Registered: 17-Dec-2008
    • Posts: 59

    I can only approximate it by rotating the bottom row, putting "Z" where "B" was, instead of on VK_102.  For Qwerty, it helps a little bit, mostly by helping to balance the load on the left ring and middle fingers with the right ones.  (It also improves "key preference", which is my subjective parameter based on how much I like using each key.)  Colemak is slightly worsened by it, further unbalancing the middle fingers (the right middle finger ends up doing about 16% of all typing, the left, 9%).  (Key preference is improved, though.)  In neither case is same-finger or two-row jumping significantly affected.

    So, my algorithm would (kinda-sorta, due to the approximation) say that it's worth it for Qwerty, but not for Colemak.

    For poops and giggles, the best single swap on Colemak in my algorithm's opinion is "O" and "S".  This reduces the load on the over-used right pinkie from ~8.5% to ~6.7% (left pinkie does ~7.4%) and increases the load on the under-used left middle finger from ~10.5% to ~12.3% (the right middle finger does ~16%).  Same-hand same-finger is increased, but "persistent same-finger" (like WORD on Colemak) is decreased.  (Personally, I think "WORD" is more awkward after the swap, but whatever).

    A comment about awkwardness--I've noticed that row changes involving the middle and ring fingers tend to feel particularly awkward to me.  This is true even of sequences like "WAS" and "FAR" on Colemak.  I think the middle and ring fingers must be bound by the same tendon or ligament, making independent movement of those fingers difficult.  Two-row jumps involving those two fingers are particularly heinous.  Maybe I should plug that into my algorithm and see what happens.

    Offline
    • 0
    • Reputation: 211
    • From: Viken, Norway
    • Registered: 13-Dec-2006
    • Posts: 5,345

    You misunderstand I believe? The fingering remains the same as before. What makes the rearrangement superior is the improved angle of the wrists and the straight stretches to the bottom row where you'd have to crook your fingers inwards before. So the loads ought to be the same for all keys except B - or do you do some modeling of the angle work?

    *** Learn Colemak in 2–5 steps with Tarmak! ***
    *** Check out my Big Bag of Keyboard Tricks for Win/Linux/TMK... ***

    Offline
    • 0
    • Reputation: 0
    • Registered: 17-Dec-2008
    • Posts: 59

    No, I don't understand.  Anyway, there's no parameter for this in my program; I didn't want to get too far away from standard keyboard usage.

    Offline
    • 0
    • Reputation: 211
    • From: Viken, Norway
    • Registered: 13-Dec-2006
    • Posts: 5,345

    But this is about standard keyboard usage as well as nonstandard. If a key is awkward to reach due to non-distance-related factors, merely weighting the distance to it is not all that interesting. And if you use too subjective weightings, the result will be too subjective as well. Since you want to model strain, you need to include more than distance and such things as  same-finger. Αnd it's not just about which row you're on either.

    I also think that 'reversals' are tricky things indeed. Sometimes they're awkward, but others can work just fine, even using the weaker fingers!

    *** Learn Colemak in 2–5 steps with Tarmak! ***
    *** Check out my Big Bag of Keyboard Tricks for Win/Linux/TMK... ***

    Offline
    • 0
    • Reputation: 0
    • Registered: 17-Dec-2008
    • Posts: 59

    Well, I added a parameter for minimizing "bad finger sequences," like QS, SE, WD, and DR on Qwerty.  It has helped quite a bit.

    I've also added a couple of Dvorak-inspired parameters that have helped a lot:  one maximizes hand alternation and the other maximizes inward typing.  Both of these things work to minimize "reversals."

    Formerly, I was getting better-feeling layouts by locking the "Qwerty keys" (ZXCV and punctuation) prior to optimization.  Now, the opposite is true.  Also, now layouts with the best overall score really are better overall!  So, my parameter set seems to have vastly improved in quality of late.  :)

    Currently, my program thinks the best single swap for Colemak is the rather unexpected A/O swap.  This really improves inward flow, but it also eliminates some "bad finger sequences" and balances the load between the pinkies a little better.  Common words like WAS and YOU become a lot easier to type.  (Another interesting example is ORANGE.)  'Course, you lose the left hand Ctrl+A.

    Offline
    • 0
    • Reputation: 1
    • From: New York, New York
    • Registered: 22-Nov-2008
    • Posts: 129

    Whoa, that does seem like a strange optimization but I don't know I'm a little afraid to actually try it out. It seems interesting that Colemak is so perfect that it is the only change that your program suggested.

    Colemak typist

    Offline
    • 0
    • Reputation: 0
    • Registered: 17-Dec-2008
    • Posts: 59

    Nah, that's just the one swap that results in the greatest improvement (according to my program).  There's a lot more it could do.

    Offline
    • 0