• You are not logged in.

Best Layouts for Hand Alternation?

  • Started by lalop
  • 41 Replies:
  • Reputation: 0
  • Registered: 04-Apr-2013
  • Posts: 538

Update: Widely Alternating Layouts

-----

I was wondering what are the best layouts with high hand alternation. 

Dvorak is the obvious example.  The reason I didn't stop there is that I have several beefs with it (l and f spots; period and, to a lesser extent, comma taking up the valuable "basically home row" qwerty we/io spots). 

The other one I've found is Klausler.  Maybe the second one in this post.

Are there any others?

Last edited by lalop (09-Mar-2014 05:49:37)
Offline
  • 0
  • Reputation: 3
  • Registered: 08-Dec-2010
  • Posts: 656

You can use Patojk keyboard analyzer and try to get your own keyboard layout with every specific preference that you would like.

Last edited by Tony_VN (30-Dec-2013 13:14:52)
Offline
  • 0
  • Reputation: 1
  • Registered: 15-Feb-2010
  • Posts: 14

Basically, most layouts with all the vowels and punctuation on one hand will have about the same hand alteration as Dvorak. The keyboard Layout Analyzer has a metric for consecutive hand.
http://patorjk.com/keyboard-layout-anal … d/nXns9FgB

Total consecutive hand (Alice in Wonderland Ch1)

Simplified Dvorak          2239
OISEAU                      2239
HIEAMTSRN                2263
Balance Twelve             2296
NRSTM                       2296
Klausler                       2312
BvoFRak EN V0.5         2372
MTGAP                       2459
QGMLWY                    2547
Vu Keys                      2509
EAton                         2547
TypeHacK Layout         2618
STNDC                       2950
Colemak                     3033

What most of these have in common is that all 6 vowels are on one hand and "most" consonants are on the other. That is achieved by adding all the punctuation on the vowel hand. STNDC seems to be the exception to the rule. Colemak is added for a reference point, not as an example of high hand alteration.

Last edited by screaminglemon (31-Dec-2013 02:59:59)
Offline
  • 0
  • Reputation: 0
  • From: London, UK
  • Registered: 09-Nov-2013
  • Posts: 270

T9-QWERTY - my port of T9 to the PC (a work in progress); T9-MOUSE - COMING SOON
Keyboard Shorthand

Offline
  • 0
  • Reputation: 0
  • Registered: 04-Apr-2013
  • Posts: 538

Thanks for the compilation, screaminglemon.  Is there a way to compare more than 5 layouts at once, or did you do all that by hand? :O

One bug/confusion in the Keyboard Layout Analyzer (KLA): when you uncheck "Instances of the same key being pressed twice in a row are not factored in", the numbers decrease, rather than increase.  I think it probably does the reverse of what it says. 

First impressions concerning layouts:

  • Klausler has good hand alternation stats, and I haven't found anything particularly wrong with it (except that comma might be less useful in programming than prose).  This one might be the best for adopting a "wide layout" (i.e. shifting the right hand by one letter right, displacing letters to the middle or above as with quotations).

  • MTGAP - another potential "wide layout", though with not-as-great hand alternation results.  My only complaint is with the w position.  Switch it and c ?

  • Balance Twelve and NRSTM seem to be identical layouts.  Seems reasonably promising.  Not sure if c/d/w should be switched to satisfy [qwerty wf/uy]'s "basically home row" positions.

  • HIEAMTSRN also seems reasonably promising.  Not sure if c/d should be switched (Update: doing this with HIEAQDTSRN increases same-finger).

  • BvoFRak is weird!  It scores well, but puts f,w,m in (IMO) very awkward bent-finger bottom-row positions, while giving z and j relatively privileged positions.

In addition to

QGMLWY                    2547

I've tried out

QGMLWB                    2486
QFMLWY                     2406

so it seems the carplax layouts do okay (though not particularly well) in hand alternation.  They probably resemble colemak more than the others, though, and can be used as wide layouts as well.  The "worst" optimization, QFMLWY, gets the most hand alternation, though we probably need better samples.

Last edited by lalop (22-Jan-2014 12:55:42)
Offline
  • 0
  • Reputation: 0
  • Registered: 04-Apr-2013
  • Posts: 538

Here's an attempted wide alternating layout based on HIEAMTSRN:

HIEAMTSRN-wide         2253

If you, like me, buy that QWERTY u is harder to press than QWERTY h, swapping them leads to:

HIEADTSRN                 2253

nhF1buU.png


All Results:

Note that the layouts are set to reflect my own touchtyping technique.

I've tried to use pseudo-workman criteria, prioritizing bottom-row letters near the middle and discouraging those nearer the sides.  An open question, though, is whether or not it's worth swapping the f out of its rather strenuous position, and if so, what the swap should be.  My best suggestion is swapping the bottom-row vwgf for wfgv, since that position seems to deal best with the OFT trigraph (you could hit the F with the pointing finger and T with the middle), but I honestly don't know.

That question notwithstanding, HIEADTSRN looks like quite an interesting layout; I'm tempted to give it a shot.  I don't have an expert eye, though, so might have missed something obvious.  Second opinion, critiques, etc welcome.

Last edited by lalop (07-Jan-2014 17:26:06)
Offline
  • 0
  • Reputation: 114
  • From: Oslo, Norway
  • Registered: 13-Dec-2006
  • Posts: 4,742
shaaniqbal said:

36 WPM after 20 days (4–8 h practice)? After that long, I'm pretty sure I did 40 WPM and Ghen/Pavel 60 with MessagEase! Meh. ;) Well, maybe a little more practice...

It looks like yet another touchscreen layout stuck in the old tap-key paradigm. On touch screens, you want to sweep as much as you tap!

Last edited by DreymaR (08-Jan-2014 16:29:41)

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

Offline
  • 0
  • Reputation: 0
  • From: London, UK
  • Registered: 09-Nov-2013
  • Posts: 270

I never suggested it was a particularly fast layout, just that it was made for alternation.

T9-QWERTY - my port of T9 to the PC (a work in progress); T9-MOUSE - COMING SOON
Keyboard Shorthand

Offline
  • 0
  • Reputation: 114
  • From: Oslo, Norway
  • Registered: 13-Dec-2006
  • Posts: 4,742

Quite. My point however, is that on a touch screen you need to think touch screen and most "new" layouts for touch screens don't quite.

If you want prediction, maybe KeyMonk is good (haven't tried it). Otherwise, MessagEase is my only recommendation and no amount of hand alternation is going to help you if your touch screen layout isn't made for touch screens! In fact, my way of using MessagEase leads to lots of thumb alternation which is pleasant (but maybe not as fast as it could've been?).

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

Offline
  • 0
  • Reputation: 0
  • From: London, UK
  • Registered: 09-Nov-2013
  • Posts: 270

A T9-MessagEase type hybrid could make for a very potent combination, I think. The dragging for letters would fill the role of unambiguous input, while an ambiguous tapping mode would be used for extremely high speed.

Swyping in the sense of drawing the outline for words is not necessarily any faster than tapping in my experience, and experimental studies I've read.

An interesting aside: The same man who was one of the inventors of T9, Cliff Kushler, also came up with Swype which was acquired by Nuance Communications, the company behind Dragon NaturallySpeaking. Voice dictation software is definitely the avenue with the highest potential for high speed input.

I think a thumb alternation layout combined with a strong autocorrect would also make for high speeds. I've explained my reasoning in an older post. I understand though that you're at odds with autocorrection.

T9-QWERTY - my port of T9 to the PC (a work in progress); T9-MOUSE - COMING SOON
Keyboard Shorthand

Offline
  • 0
  • Reputation: 0
  • Registered: 04-Apr-2013
  • Posts: 538

Version two of the layout.

Miz8MIJ.png

HIEAQDTSRN   2236 

Results: http://patorjk.com/keyboard-layout-anal … d/FvpqsKzJ

Most of the changes since last time were eyeballed and judgement called, with benchmark checking only to make sure they don't sacrifice too much.  Clearly, the most interesting aspect is the j/x placement.  IMO, the z/q positions are the hardest to reach, followed by those.

The period position (rather than at qwerty v) is to allow for easier typing of the .' digraph: . can be typed with the index finger, then ' with the middle.  Arguably, however, this might be too wasteful, and the period stretch might make subsequent pressing of space by the left hand more difficult.

Although w is more common than g, g is often double-tapped, and so gets the easier position for doing so.

I sort of dislike the parenthesis position.  It may be too heavy a compromise.

As always, suggestions, critiques, observations I might've missed, are appreciated.

I think I'll be alpha testing this as a secondary layout (primary will have to wait at least until I set up emacs to be layout-neutral) and making further changes on the fly.

Last edited by lalop (20-Jan-2014 11:46:18)
Offline
  • 0
  • Reputation: 0
  • Registered: 04-Apr-2013
  • Posts: 538

It is a bit mad.  MTGAP came up with those positions for z/q, but in this case there was a harder position to put those (wide layouts are problematic in that there's not as much space to the right to put keys; instead, they go to much more difficult positions in the middle). 

I'll have to see how it pans out in practice.  If it doesn't, might have to revert to a HIEADTSRN-like layout.

Edit: after a small amount of typing, the . position is quite annoying - I've swapped them.   Maybe using the thumb to hit the . and index to hit the ' would work on that digraph instead.

Edit2: With the current m/d swap, the nd digraph is a bit of a stretch. Not sure if that's enough to warrant moving it back.

Edit 3: reswapped m/d back, leaving HIEAQMTSRN.  I have to admit, in addition to nd and ld feeling less weird, rolls at that d position are pretty nice, including "could", "would".

j/x/z/q are rare enough I barely have occasion to use them, going as slowly as I've been.  I use b/p more than I'm comfortable with, though; might have to start getting used to pinkying them.  Colemak is a nice layout for not having to strain the pinkies to those spots.

The layout reminds me of QWERTY, strangely enough, in how it doesn't try all as hard as colemak to stay by the home row.  There are occasional times when a word would move to, then stay around the top row, and in those times it's better to keep that alternative fingering rather than to try to home-row touchtype, again much like QWERTY and unlike colemak.

Edit 4: tempted to swap l and c.  It seems to makes the stats only negligibly worse, while distributing more work to the strong middle finger.  Worried about awkward transitions (e.g. "fal")

Last edited by lalop (23-Jan-2014 15:27:03)
Offline
  • 0
  • Reputation: 0
  • Registered: 04-Apr-2013
  • Posts: 538

To get more detailed analysis about hand-alternation, I wrote a script to compute (and slightly expand on) carpalx's rh(n) = P(alt w/i n), "the probability of hand alternation within n letters".  P(alt w/i 1) is just the usual hand-alternation rate. 

P(alt w/i n) \ n      1     2     3     4     5     6     7     8     9    10
HIEAQMTSRN:        80.2  96.6  99.2  99.8  99.9 100.0 100.0 100.0 100.0 100.0
Dvorak:            79.9  96.1  99.2  99.8  99.9 100.0 100.0 100.0 100.0 100.0
MTGAP:             77.0  94.6  98.5  99.6  99.9 100.0 100.0 100.0 100.0 100.0
QGMLWB:            77.2  93.9  97.7  98.9  99.4  99.7  99.8  99.9 100.0 100.0
Wide-Colemak:      71.6  90.0  96.1  98.3  99.2  99.6  99.8  99.9  99.9 100.0
QWERTY:            69.0  87.6  95.4  98.1  99.2  99.6  99.8  99.9 100.0 100.0

The complement can be seen as the percentage of (n+1)-sized strings that are same-hand:

P(n+1 same hand)\n+1      2     3     4     5     6     7     8     9    10    11
HIEAQMTSRN:           19.77  3.41  0.84  0.23  0.06  0.02  0.01  0.00  0.00  0.00
Dvorak:               20.06  3.85  0.81  0.17  0.06  0.02  0.01  0.01  0.00  0.00
MTGAP:                23.02  5.35  1.47  0.39  0.11  0.04  0.02  0.01  0.00  0.00
QGMLWB:               22.85  6.11  2.32  1.14  0.57  0.29  0.15  0.08  0.04  0.02
Wide-Colemak:         28.43  9.98  3.85  1.69  0.83  0.43  0.22  0.11  0.06  0.03
QWERTY:               30.99 12.35  4.65  1.89  0.82  0.40  0.18  0.08  0.04  0.02

Taking the recriprocal, you can expect to get a same-hand string of length l every this many characters:

E(same length l)=1\l      2     3     4     5     6     7     8 
HIEAQMTSRN:               5    29   119   427  1547  4369 10609
Dvorak:                   5    26   123   580  1768  4369  7427
MTGAP:                    4    19    68   254   928  2856  5713
QGMLWB:                   4    16    43    88   176   347   657
Wide-Colemak:             4    10    26    59   120   230   445
QWERTY:                   3     8    22    53   121   251   563

Clearly, small percentages can make a pretty big difference in how often you'd expect to see a long chain.  For example, this paragraph is 422 characters long, so we would expect to see a same-hand string of length 8 in colemak!  This is likely the reason certain layouts "feel" like they are low-alternation, despite seemingly okay P(alt w/i 1) and P(alt w/i 2) numbers.  The usual n=1 analysis misses out on quite a lot.

By contrast (and as seen in the examples below), the HIEAQMTSRN combos pretty much stop after n=4 (i.e. same-hand strings of length 5), and you would expect to see only one such string in the above paragraph.


These numbers were generated under the following options:

  1. Spaces don't belong to either hand (so they always interrupt a hand combo). If space is set to one hand, P(alt w/i 1) generally becomes around 20% worse.

  2. Letters that are repeated (e.g. "jjj") are compressed into one letter ("j")

  3. Unknown characters in the input are treated as not belonging to either hand, similarly to space.

You can come up with your own stats with different inputs, or change up the function call arguments in the code.  To add a new layout, modify the layouts_hands variable.  Usage is:

python alternation.py TEXT-INPUT-FILE


Another utility of the script is to extract examples of same-hand combos:

QWERTY (length 11):
exaggerated


HIEAQMTSRN (length 9):
(~1%/1%).

Wide-Colemak (length 9):
henomenon

QWERTY (length 9):
essagease
scattered
-jump-lin
aggregate
regretted


HIEAQMTSRN (length 8):
(~2%/2%)

Wide-Colemak (length 8):
paragrap
one-line
omeone?"

QWERTY (length 8):
eartbeat
terested
defeater
stressed
stracted
monopoli


Dvorak (length 7):
equippe
"Okay,"

Wide-Colemak (length 7):
imminen
'mine',
enjoyin
bastard
mnemoni
onome?"
million
eyeline

QWERTY (length 7):
ecessar
serbase
garbage
bserved
streets
watered
attress
creased
bserver
terface
bastard
grabbed
deserve
stagger
referab
screwed
aracter
started
crested
million
strateg
strated
garages
-jumpin
reacted
ggested
dressed
opinion
treated
dragged


Dvorak (length 6):
eappea
okay."

Wide-Colemak (length 6):
enemy,
line?"
likely
"I'm-"
ominou
scratc
ullen.
minimi
enemie
you'll
elimin
ull-on
"Hey."
garbag
olley.
trasts
homele
money,
Lookin
[swapp
illy."
"Huh,"
ommuni
hool."
partwa
"Hey,"
lookin
-line.
stract
hinkin
home,"

QWERTY (length 6):
Regard
ceases
argest
dersta
better
ull-on
njoyin
affect
illy."
Agreed
refres
trasts
atever
ggests
"I'm-"
scratc
sweats
ommuni
stretc
access
hinkin
etwatc
betwee
stared
excess
hool."
lookin
fester
ftware
ffered
minimi
ominou
etters
swered
versat
rrect!
Waterc
Lookin
"Huh,"
rected
estate
bestse
carved
scated
earest
effect
rdered
severe
assets
berate
you'll
severa


HIEAQMTSRN (length 5):
eyboa
abe-a
(i.e.
y-yea
aybe,
nts?"
Yeah,
"pwd"
(1.3%
llsgl

Dvorak (length 5):
ookie
ay'."
poke,
ngths
you."
"You'
ake."
you,"
ake,"
pooke
key".
"I...
n-thr
appy.
appy,
eppy.
kup,"
ake!"
ghts-
okay,
chscr
nd-ch
appoi
you!"
llsgl

Wide-Colemak (length 5):
hile,
looke
lemen
ion."
hooke
le...
innin
ion,"
mile.
parts
scatt
"Look
"Mm,"
key".
ookie
joine
look,
drast
aract
pract
emon,
crast
appar
twatc
attac
illin
uel."
o-one
ollin
"'Lo,
omen,
(qwfp
enien
sfact
Swapp
eelin
ink,"
grabb
illne
stats
ike?"
"No,"
scrap
ionin
NONIN
ellin
ene,"
omly,
mily'
ly...
mily,
mily.
uinin
onomi
spraw
hill,
-Hole
inkie
stagg
huh,"
oken.
hem."
likin
mine.
unnin
melle
TRACT
e-jum
apart
rward
scarf
ullie
nyone
strat
nnon?
garag
ully,
ftwar
omeho
you?"
you."
olume
start
helme
him."
home.
lone,
"I'll
dragg
sadva
ell,"
milin
home-
scars
mely:
hool?
hione
ill."
hink.
arst)
"You'
ommen
hollo
hole.
you,"
line,
hool,
"I...
hone.
swadd
inuin
inion
he'll
momen
"No."
-home
barbs
WARRA
aggra

QWERTY (length 5):
strea
innin
ion,"
"Look
illin
-pop.
ollin
impli
ly...
mily'
mily,
mily.
safeg
terva
milin
great
likin
eeded
vergr
immin
swers
nipin
Tears
ully,
cceed
eeves
you."
secat
kup,"
IMPLI
eezed
hool?
sadva
ereas
hool,
hink.
-up?"
tages
after
inuin
barbs
refra
wease
cerea
exact
esser
gered
look,
breat
crast
"Mm,"
pillo
tterb
eter!
derat
uinin
onomi
you?"
eared
scars
earer
ionin
scarf
artwa
ceeds
"You'
ion."
scree
drast
added
swadd
"Lump
stats
"No."
certa
tears
NONIN
WEVER
trees
areas
Impul
"'Lo,
stead
vered
estab
omly,
bsess
garde
sfact
wever
huh,"
tated
ppoin
unnin
dated
faces
sweet
caref
ffers
serab
pinki
you,"
"I...
edged
TRACT
arder
philo
reass
aggra
gates
"I'll
testa
sages
hill,
ink,"
WARRA
attac
eaves
swear
rward
nnon?
"No,"
him."
Certa
limin
ill."
oppin
descr
hollo
aster
astes

To create some, add some lengths to the list examples_lengths.


Update:
27 Jan version - huge rework: added data structures, improved table padding & printing, example filtering no longer case sensitive
29 Jan version - updated example generation

Last edited by lalop (29-Jan-2014 13:45:47)
Offline
  • 0
  • Reputation: 0
  • Registered: 04-Apr-2013
  • Posts: 538

Version 3 of the layout is to revert my attempted swaps.  The m/d swap didn't work out due to the common nd digraph, and the g/w swap due to the many letters beginning with w.  I tried, afterwards, to move the d to a better position, but each rearrangement I tried came with some serious drawbacks.  The original HIEAMTSRN was pretty darn well-considered.

HIEAQMTSRN      2236
FBTCJUg.png

HIEAMTSRN-wide  2254
7R3THKV.png

Results: http://patorjk.com/keyboard-layout-anal … d/tF9HH4ss


With HIEAQMTSRN, I've found myself forced to [learn to] use the pinkies on their top-row spots, unlike my habit with colemak.  As such, I've changed that setting in all relevant KLA layouts.

The " spot is still tentative.  KLA seems to think same-finger significantly improved if it's swapped with - (to QWERTY v), but I'm not sure if that's an artifact.

After playing with it for a few days (though being careful to overpractice colemak so as to - hopefully - avoid displacing it), I'm at around 5-10wpm.  The feel is very different; I notice it most when I go back to colemak.  HIEAQMTSRN seems to benefit more from non-standard fingering (mainly flowing up and down the top and middle rows), unlike colemak which is mainly restricted to the home row with occasional exceptions.

I've noticed more weird contortions in HIEAQMTSRN, the worst offender so far being fl.  I should probably come back to the home row before reaching for the l.

A common theme I've noticed, both in looking at and trying to modify alternating layouts, is letters being forced in "hard" bottom-row positions and top-row pinkie/stretch positions, simply because there's nowhere else to put them.  Solving this issue without sacrificing too much alternation (as well as preferably little or no same-finger) seems another fruitful direction to take this.

Last edited by lalop (29-Jan-2014 16:21:29)
Offline
  • 0
  • Reputation: 2
  • Registered: 25-Oct-2013
  • Posts: 136

@lalop, how do you design the layouts? mtgap, carpalx? tweaked to your own insights? or manually change layouts (for instance in patorjk's app), and/or try them out in real life (the best test).

There's not much scientific evidence on alternating/ rolling, right? mtgap's done some study, but the best way would be in my humble opinion:
- design unfamiliar rolling, alternating and "in between" keyboards
- give uniform groups of volunteers a training in these keyboards
- have them type test texts on the keyboards
- measure speed, errors, muscle tension (=objective data)
- ask them to report how fast/pleasant the keyboard felt (=subjective, self-reported data).

Agree?

Offline
  • 0
  • Reputation: 0
  • Registered: 04-Apr-2013
  • Posts: 538
pieter said:

@lalop, how do you design the layouts? mtgap, carpalx? tweaked to your own insights? or manually change layouts (for instance in patorjk's app), and/or try them out in real life (the best test).

All of the above.  The final step is always manual tweaking (and often a lot of experimentation), though.

pieter said:

uniform groups of volunteers

:(

Anyway, I think such tests would have to be on a finer level than keyboard layout - there are too many variables in layout design, and I doubt one could simply come up with layouts that are substantively similar aside from alternation/rolling.  Rather, it should probably be done on the corpus level (simply having the user type rolls and/or alternations), which is good in that it doesn't require learning a new layout, but may not be representative of actual text.  MTGAP's natural experiment seems like a decent compromise for the latter.

Offline
  • 0
  • Reputation: 2
  • Registered: 25-Oct-2013
  • Posts: 136

The type of switches in the keyboard matter as well....  I'd say. Hand alternation is welcome when the switches requiere much force - like the old day typewriters. When you use light switches, rolls get easier and attractive. Not to the point of wanting to type whole sentences on one hand of course, but the optimal hand alternation change might rise from, say, 35%, to 60%.  So, in that thinking, a keyboard with MX Browns may 'want' more rolls, whereas a MX Greens keyboard wants more alternation..... right?

Offline
  • 0
  • Reputation: 0
  • Registered: 04-Apr-2013
  • Posts: 538

I think that's a reasonable hypothesis (though I don't think there's an "optimal hand alternation" per se).

Incidentally, I've been using reds for a few weeks, and don't mind the same-hand stuff nearly as much as on rubber domes. starcraft minimum unhook.

Offline
  • 0
  • Reputation: 2
  • Registered: 25-Oct-2013
  • Posts: 136

I'm using my own Juli layout now:
. u o p y x c l b v
a i e n h m d r t s
: , ? k q f g w j z

I saw the other day that it is remarkably like Dvorak:
‘ , . p y f g c r l
a o e u i d h t n s
; q j k x b w m v z

Learning is hell.....

Last edited by pieter (22-Jul-2014 15:24:48)
Offline
  • 0
  • Reputation: 114
  • From: Oslo, Norway
  • Registered: 13-Dec-2006
  • Posts: 4,742

Indeed, it does look a lot like Dvorak. But you say it scores a lot better for the languages you tested with patorjk's tool? Or marginally better?

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

Offline
  • 0
  • Reputation: 2
  • Registered: 25-Oct-2013
  • Posts: 136

@Dreymar: comparison of Juli16, Dvorak, Colemak and Qwerty; using Dutch text and English text. Take the Patorjk's results for what they are (I know they don't account for good rolls. Colemak and Juli16 are designed with both rolls and alternation in mind. So, that being said, the results:

DUTCH
Rank + score (higher = better)
1. Juli16 70.18
2. Simplified Dvorak 67.16
3. Colemak 65.35
4. QWERTY 50.11

Distances (less = better)
1. Juli16 88,481.1
2. Colemak 91,814.9
3. Simplified Dvorak 105,692.4
4. QWERTY 170,998.2

Finger use The distance (!) each finger travels. total travel of all fingers = 100%. Thismeasure is more precise than number of keystrokes, because it takes in account that for instance Colemak D and H are further away than T and N, thus more taxing.
JVxZ8yd.png

Consecutive Hand and Thumb Use - How often the same hand was used to type a key as was used to type the previous key (thumbs are grouped as a separate entity). An example of this would be typing "af" on QWERTY. When looking at "f", the program notes that the left hand was also previously used to type "a". Lower = more alternation. The optimum is subjective - Dvorak wants lots of alternation, Colemak wants less alternation but more rolls. Juli16 sits sort of in between these two. So here we go:
Consecutive Hand and Thumb Use
1.Simplified Dvorak 11,073
2.Juli16 14,329
3.Colemak 18,539
4.QWERTY 20,667


ENGLISH
Rank + score (higher = better)
1. Juli16 65.32
2. Simplified Dvorak 64.79
3. Colemak 64.51
4. QWERTY 50.11

Distances (less = better)
1. Colemak 105,369.5
2 Juli16 109,852.9
3. Simplified Dvorak 110,764.0
4. QWERTY 182,733.2

Finger use The distance (!) each finger travels. total travel of all fingers = 100%.
A7dn9wE.png

Consecutive Hand and Thumb Use Lower = more alternation.
1. Simplified Dvorak 13,234
2. Juli16 15,700
3. Colemak 18,559
4. QWERTY 19,795


CONCLUSIONS - part A
- Again, based on the not-too-sophisticated Patorjk-statistics!
- I tested here only Dutch and English, due to time constraints, real life stuff
- In English and Spanish the layouts Balance12 and HIEAMTSRN score even better than Juli16; for Dutch and several other languages Juli16 is the "king" ;-)
- don't take all this too serious, shall we? messing with keyboard layouts is an utterly irrelevant micro-optimization hobby for a subset of geeks. Do we all agree? ;-)  So it's not win or lose :-) But - Juli16 is the King of course ;-)   

- all layouts score much better than Qwerty. If you are in doubt, any of these fine layouts (Colemak, Dvorak and my own Juli16) will give you a way nicer typing experience. Colemak and Dvorak have the advantage of being often pre-installed and have way more support. In a way we are nitpicking here. Am I overoptimizing? Could well be ! On the other hand, if I must learn a new keyboard layout, why not go for the optimum ? There is a lot to say both for 'satisficing', that is, pick a reasonably good layout (i.e. either Colemak or Dvorak, try both and pick the one that feels best) and live with it; but... there is also a lot to say for optimizing!  OK. Enough of philosophy, let's get down to the naked facts :) 

- In terms of hand alternation, Juli16 sits in between Colemak (less alternating) and Dvorak (more alternating). Whether or not this is good, is subjective.

- in terms of pure distance, Colemak scores best in English text, whereas Juli16 is the number 1 in Dutch. Both outcomes were to be expected.

- in terms of how the travel is distributed over fingers, Juli16 scores best, in both languages. But the difference is larger in Dutch, where Colemak and to a lesser extent Dvorak don't do so well. The reason is in the letter frequencies.

Dutch order
enati rodsl gvhkm ubpwj zcfxy q

etaoi nshrd lcumw fgypb vkjxq z
English order 

You see that in Dutch, the letters that are typed with the index fingers are more frequent: g l v k b p j. This means that these relatively hard to reach letters get typed quite often, which loads the index fingers. In other words: these letters are in too bad spots for Dutch, while letters such as c, q, y, f and u,  are in "too easy"  spots.  For Dvorak, roughly the same holds true. If in Dvorak I swap D-H,  and U-I, the results are better.

CONCLUSIONS - part B
- Juli16 versus Dvorak: In English the difference is marginal. If I typed mainly English and were to chose between Dvorak and Juli16, I'd chose Dvorak. But for Dutch there is a clear difference. For Dutch I would definately chose Juli16.

- Juli16 versus Colemak : Juli16 ranks higher than both Dvorak and Colemak in English, but to be honest these differences are small. In reality it is safe to say that Colemak, Dvorak and Juli16 are equals for the English language. For English, I'd choose Colemak (or Dvorak), for practical reasons. Pick the one that feels best. For Dutch, the difference is larger. Juli16 is better than Colemak for Dutch.

Last edited by pieter (22-Jul-2014 23:10:44)
Offline
  • 0
  • Reputation: 0
  • Registered: 04-Apr-2013
  • Posts: 538

You should include "Consecutive Finger Use" (aka same-finger), since that's easily one of the most annoying stats. (Personally, I'd deselect "Include instances of the same key..." since double-tapping is, IMO, a very different experience from having to actually move then re-tap the same-finger, but I guess it wouldn't hurt to include both.)

Also, including the lengths of your corpi will allow the alternation and same-finger rates to be converted into percentages.

Offline
  • 0
  • Reputation: 114
  • From: Oslo, Norway
  • Registered: 13-Dec-2006
  • Posts: 4,742

Thanks for the info. I wouldn't say that either optimized contestant is any better for English from those results, no – but for Dutch the difference looks substantial.

Congratulations, then, on giving birth to the king – of Holland. ;)

As Lalop says, same-finger is important. It's included in the final score if I remember correctly, but the partial score is also interesting.

Last edited by DreymaR (23-Jul-2014 10:37:10)

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

Offline
  • 0
  • Reputation: 2
  • Registered: 25-Oct-2013
  • Posts: 136

Thank you lalop, and sorry for hijacking your thread - although it IS related and possibly inspiring :-) I agree with you on the consecutive finger stats, let's see if I find the time.

Juli16 is a bit like Dvorak, but has more rolls and less alternation. The interesting thing is, I have rolls on the top and bottom rows as well:
you/yo/oy ou/uo op/ po cl/bl   I see that the HIEAxx layouts have rolls on the toprow as well. BTW, the 'Viminitz layouts'  HIEAxx and Balance12 do very well for English and also for several other languages. To me, those are the best layouts, together with Colemak, mtgap's layouts, my own layout and Dvorak.

In the end a layout is personal thing. patorjk, michael 'mtgap', andong 'ddvorak' and martin 'carpalx' all have made wonderful algorithms. But do you agree with the weights, penalties and bonusses?

For on thing, mtgap sees inrolls as good (agree) but puts a penalty on outrolls. I don't completely agree, I guess it depend on your fingers. My index is much shorter than my middle finger.  My best rolls are (from good to mediocre, low number are good):
1. middle-ring = outward roll
2. ring-middle = inward roll
3. middle-index = inward roll
4. index-middle  = outward roll
5. pinky-ring  = inward roll
6. ring-pinky  = outward roll

Also, not all "same finger combos" are equal. Qwerty ED or WS are pleasant, Qwerty FT is bad. 

Regarding optimal alternation, I tend to agree with 'michael mtgap' that 2 left, 2 right may be optimal. BTW, I don't fully understand the hand switch percentages in patorjk, I can't translate them to a Bayesian statistic.... Meaning, that if I have a sequence of 4 keystrokes, these are the possibilities (L=left hand, R = right hand):
LLLL bad
RLLL ok
LRLL ok
LLRL ok
LLLR ok
RRLL good!
LRRL ok
LLRR good!
RLRL ok, maximum alternation, too much?
LRLR ok, maximum alternation, too much?
RLLR ok
LRRL ok
LRRR ok
RLRR ok
RRLR ok
RRRL ok
RRRR bad

If the chances of a consequtive keystroke with the same hand are 100%, you d have basically a one hand layout (all LLLLLLL or RRRRRRR). If the chances are 0%, you would have a maximum alternation: LRLRLRLR.  But what percentage goes with the ideal of LLRRLLRRLLRR ???

Offline
  • 0
  • Reputation: 2
  • Registered: 25-Oct-2013
  • Posts: 136

By the way, an interesting side of my layout is that 29 keys changed place, compared to qwerty. Only the A stayed in its place. And you know what? I think it doesnt matter: learning a new layout is hard, no matter what.....

Edit: and regarding the best alternation, I am now in the 2x2 camp, also known as RR LL RR LL. Based on my own (N=1) feeling.

Last edited by pieter (26-Jul-2014 20:56:48)
Offline
  • 0