Hello, I imagine you must get lots of e-mails regarding the layouts you have proposed and the statistics used to measure their performance, so forgive me if I'm just adding onto a long list of unwanted e-mails. First off, I would like to preface this by saying that I have gone from QWERTY to Dvorak to Colemak and was able to perform at a respectable enough speed in each in order to appreciate their advantages over each other. I've now begun to attempt to switch over to QGMLWB out of curiosity, not so much for the decrease in effort, but to experience the "feel" of the layout. For example, though Dvorak and Colemak are not that different in effort, they have very different feels to them in the alternation versus rolling motion approaches and though I had similar typing speeds in each, I greatly prefer Colemak due to the feel of it. I have only been typing with QGMLWB for a few days now and feel that it has a nice combination of both strategies. I'm not sure if you go beyond the analysis of the simulation to try to understand why the converging solution is superior, but I would like to note that the inclusion of the letter H on the right hand with the vowels as opposed to in Dvorak where the left hand is simply vowels is rather interesting. Though you lose the TH digraph on the right hand that you would have in Dvorak, which is the most common English digraph, you gain HE which is the second most and also stand to gain all the other H-vowel digraphs. For example, TH often occur in words such as "the," "that," and "though," but in each of these cases you still maintain an H-vowel digraph. What I was more curious about is why the left hand has STN in that order. A look at common digraphs would make me think that having the N before the T (as is done in Dvorak) would be much more useful in terms of performing rolling motions (though perhaps it would unbalance finger load, which is something more difficult for me to analyse). Furthermore, I would even think that having NST in that order would be even more useful as S and T seem to have a tendency to follow N rather than leading it, not to mention the use of the letter S to pluralize any word that may end in N. Also, as T is a more common letter than N, moving it as such would put it on the more apt index finger (though again, perhaps this is undesirable as the index finger is already busy reaching for the middle column). One thing that is influencing it is perhaps the fact that T and L are on different fingers in the STN arrangement why would reduce same finger usage for words ending with -tly whereas in NST they would lie on the same finger, though perhaps M and L could be switched in order to accommodate this? Now I fear that perhaps I have gone too far. I've downloaded carpalX in the interest of trying out this proposition myself, though I'm afraid that trying to determine how to use it is outside of my expertise. In any case, I was curious if you had a moment to spare a thought regarding this. This is a site I was looking at regarding the frequency of letters/digraphs, which I can only assume to be relatively accurate: http://scottbryce.com/cryptograms/stats.htm Thank you for your time, Stephen O'Connor ### ### ### Hi Stephen, I very much appreciate your thoughts. I'm on QWERTY and have approached the layout problem rather theoretically, more than practically. I am hopeful to leave the front-line work to users :o Oops, did I say that? The QGMLWB layout was determined in an automated way. There was no tweaking by hand. The position of keys is a reflection of the rules that went into defining effort and the subsequent attempt to minimize this effort. Something like 'that' works pretty well on QGMLWB - it's D;JD on QWERTY which is a nice right-to-left rolling motion on home row. In Colemak, there is no three-stroke roll. The rules are made up of several components, so I cannot point to a single one to explain why STN appears as it does. It is likely that the path component of the effort (where I try to penalize combinations that skip rows, use the same finger and favour rolling combinations) had a significant effect on this position. Carpalx works on the basis of trigraphs to allow for longer-range impact on effort (e.g. zzz can be defined as much more difficult than zz). Finger penalties will also impact the layout. I include some detailed statistics below. STN has uniformly better statistics, but only slightly. These statistics capture some aspects of typing, but not all. Given the small differences between STN and NST, practical experience may indicate that NST is better. Ideally, it would be great to operationalize this so that it can be added to the statistics and underlying rules. One can use carpalx to evaluate a keyboard's efficiency, given any layout. I've attached the full effort profile for QGMLWB and two modifications: NST and TNS, in place of STN. These at first may appear complicated, but if you read http://mkweb.bcgsc.ca/carpalx/?keyboard_evaluation and then see the formatted QGMLWB report http://mkweb.bcgsc.ca/carpalx/?keyboard_layouts you can make sense of the format. In the code, tutorial-01.conf shows how to evaluate a keyboard (QWERTY is used). Below, I'll call QGMLWB=STN 1. Total effort comparison b = base (finger distance) p = total penalty (ph - hand, pr - row, pf - finger) s = path STN Keyboard effort ------------------------------------------------------------ b 0.382 22.9 22.9 p 0.570 34.2 166.9 ph 0.000 0.0 0.0 pr 0.153 26.8 26.8 pf 0.363 63.6 90.5 s 0.716 42.9 100.0 all 1.668 100.0 100.0 NST Keyboard effort ------------------------------------------------------------ b 0.382 22.3 22.3 p 0.587 34.3 165.0 ph 0.000 0.0 0.0 pr 0.153 26.0 26.0 pf 0.379 64.5 90.6 s 0.741 43.3 100.0 all 1.710 100.0 100.0 TNS Keyboard effort ------------------------------------------------------------ b 0.382 22.0 22.0 p 0.629 36.2 160.7 ph 0.000 0.0 0.0 pr 0.153 24.3 24.3 pf 0.419 66.6 90.9 s 0.726 41.8 100.0 all 1.737 100.0 100.0 The base effort is the same because rotating STN does not change total finger distance. The difference between STN and NST is in the penalty and stroke. The finger penalty is higher (.379 vs .363) and the stroke penalty is higher (.741 vs .716). This is a rather abstract difference, since it's not clear how these values impact the typing experience. Below, I'll focus on STN vs NST 2. finger frequency STN keyboard finger frequency ------------------------------------------------------------ 0 415916 4.5 4.5 1 762974 8.3 12.9 *** 2 1291809 14.1 27.0 *** 3 1959774 21.4 48.4 6 1954935 21.4 69.8 7 1384856 15.1 84.9 8 801545 8.8 93.7 9 579356 6.3 100.0 NST keyboard finger frequency ------------------------------------------------------------ 0 415916 4.5 4.5 1 816882 8.9 13.5 *** 2 1034612 11.3 24.8 *** 3 2163063 23.6 48.4 6 1954935 21.4 69.8 7 1384856 15.1 84.9 8 801545 8.8 93.7 9 579356 6.3 100.0 NST loads the ring finger more (13.5% vs 12.9%), but lowers use of middle finger (24.8% vs 27.0%). I treat the middle finger as stronger than index, and this is why greater use of index is penalized. Colemak keyboard finger frequency ------------------------------------------------------------ 0 764404 8.4 8.4 1 763284 8.3 16.7 2 988856 10.8 27.5 3 1798350 19.7 47.2 6 1909344 20.9 68.0 7 1384856 15.1 83.1 8 829340 9.1 92.2 9 712731 7.8 100.0 STN middle finger use is very similar to colemak (STN 27% Colemak 27.5%). 3. Finger run length - left hand STN 1 8001149 91.3 91.3 2 705057 8.0 99.4 3 50113 0.6 99.9 4 5066 0.1 100.0 5 639 0.0 100.0 6 71 0.0 100.0 7 12 0.0 100.0 8 5 0.0 100.0 9 7 0.0 100.0 10 1 0.0 100.0 11 4 0.0 100.0 12 1 0.0 100.0 16 1 0.0 100.0 17 1 0.0 100.0 20 1 0.0 100.0 NST 1 7660396 89.3 89.3 2 847942 9.9 99.1 3 68061 0.8 99.9 4 5349 0.1 100.0 5 641 0.0 100.0 6 70 0.0 100.0 7 13 0.0 100.0 8 4 0.0 100.0 9 3 0.0 100.0 10 5 0.0 100.0 11 4 0.0 100.0 12 1 0.0 100.0 16 1 0.0 100.0 17 1 0.0 100.0 20 1 0.0 100.0 The run length is the number of successive strokes with the same finger. STN has 91.3% of successive keys hit with different fingers and NST slightly lower at 89.3%. ### ### ### Firstly I just want to say thank you for taking the time to reply and thank you for providing the statistics regarding my suggestions. I want to say that I am very inspired by your work but at the same time, as you described in your studies, that a large portion of determining which layouts are optimum is a matter of deciding what parameters to test and how to evaluate desired criteria within these parameters. While I am fascinated at the number for QGMLWB's reduced effort with a sort of statistical reverence, I do think that there are other human factors that are very worth considering that are not accounted for. This is not to belittle your work, as my opinions of it are quite the opposite and I love the detail to which you have gone. However, other studies, notably those done on the "Workman" layout (http://viralintrospection.wordpress.com/2010/09/06/a-different-philosophy-in-designing-keyboard-layouts/) have made me think of keyboard layouts in a very different way. While I have issues with the Workman layout as well, it is the human aspect of his reasoning that interest me, particularly the realization that some fingers prefer to extend while others prefer to curl. More specifically, the ring and middle fingers are more suitable to extending while the index finger is more suitable for curling. This realization has changed my perspective regarding the penalties typically attributed to the top and bottom rows. For instance, I consider the Q and P positions in QWERTY to be worse than Z and / as the pinky may curl to these positions without disrupting the rest of the hand (though it is still far from comfortable). The upper row pinky keys on the other hand require the wrist to be adjusted in order to press them. Along the same line of thought, I find that the Y position is worse than the B position, as again in this case moving to the upper row requires that the whole hand move whereas for the bottom row this is not the case. I would invite you to try this simple motions in person as performing while considering these kind of statements is what truly made me realize how much truth they held. These realizations also have other impacts, as they affect the ease of certain rolling motions that would otherwise appear to be quite difficult. All this has led me to attempt to create a layout of my own, though I still have much to consider about it. Perhaps I will share it with you once I am satisfied with it and have tested it through personal experience. For the moment I merely wished to express my thanks once more and share with you these insights in the hopes that you find them interesting. While I likely cannot determine a layout that has such statistically low effort, I do believe that there are potentially layouts that are more suitable to the form of the human hand using factors that have not been fully considered. Another factor that I sometimes consider is the use of a grid layout keyboard as I own a Kinesis Contoured Advantage (http://www.kinesis-ergo.com/advantage.htm) which changes some things, though when pondering my own layouts I do so with a staggered keyset in mind. I was somewhat curious as to what the results would be if the comma, period and forward slash keys were not necessarily held still. Much of my though process regarding creating my own layout is inspired by considering the layouts proposed by others; imitation is the sincerest form of flattery after all. Also, do you not feel curious to try your own optimized layout? ;) ### ### ### Hi Stephen, I've added the workman layout to the carpalx keyboard layout page and made mention of our conversation on the front page in a new section "statistics vs human factors". I am in absolute agreement that human factors are critical in keyboard layout design. The stats should only be revered if they faithfully capture both essentials and nuances of typing ergonomics. > More specifically, the ring and middle fingers are more suitable to > extending while the index finger is more suitable for curling. Interesting point, and I agree. This could be incorporated into the simulation by adjusting the finger travel distance, so that, if I understand you correctly, the perceived distance between s-w is shorter than s-x (because ring finger likes to extend) and f-r is greater than f-v (because index likes to curl). 0 I ran the workman layout through the corpus on which the results I sent previously were based on. ### effort # workman b 0.336 16.9 16.9 p 0.848 42.5 139.7 ph 0.000 0.0 0.0 pr 0.200 23.6 23.6 pf 0.505 59.6 83.2 s 0.809 40.6 100.0 all 1.993 100.0 100.0 # qgmlwb b 0.382 22.9 22.9 p 0.570 34.2 166.9 ph 0.000 0.0 0.0 pr 0.153 26.8 26.8 pf 0.363 63.6 90.5 s 0.716 42.9 100.0 all 1.668 100.0 100.0 Your base effort is much lower than qgmlwb (.336 vs .382) but row/finger penalty is much higher (.848 vs .570). The penalty difference can be accounted by my definition of penalty, which doesn't distinguish curl/extension like you mention. ### row frequency # workman 1 1946938 21.3 21.3 2 6178969 67.5 88.8 3 1025258 11.2 100.0 # qgmlwb 1 1731346 18.9 18.9 2 6734080 73.6 92.5 3 685739 7.5 100.0 Perhaps I attempt to load the home row too much? ### finger frequency # workman 0 764404 8.4 8.4 1 980016 10.7 19.1 2 1349950 14.8 33.8 3 1689762 18.5 52.3 6 1470419 16.1 68.3 7 1384856 15.1 83.5 8 868605 9.5 93.0 9 643153 7.0 100.0 # qgmlwb 0 415916 4.5 4.5 1 762974 8.3 12.9 2 1291809 14.1 27.0 3 1959774 21.4 48.4 6 1954935 21.4 69.8 7 1384856 15.1 84.9 8 801545 8.8 93.7 9 579356 6.3 100.0 My left pinky use is ~50% of yours. Right pinky use is more-or-less the same. I load the ring fingers less than you do and favour index. Again, this is a penalty thing and may be too extreme on my part. ### hand frequency # workman 0 4784132 52.3 52.3 1 4367033 47.7 100.0 # qgmlwb 0 4430473 48.4 48.4 1 4720692 51.6 100.0 Our hand asymmetry is nearly the same. ### finger run length # workman 1 8255001 92.8 92.8 2 605437 6.8 99.6 3 35889 0.4 100.0 4 2675 0.0 100.0 5 216 0.0 100.0 6 47 0.0 100.0 7 6 0.0 100.0 8 1 0.0 100.0 9 5 0.0 100.0 10 1 0.0 100.0 11 3 0.0 100.0 12 1 0.0 100.0 # qgmlwb 1 8001149 91.3 91.3 2 705057 8.0 99.4 3 50113 0.6 99.9 4 5066 0.1 100.0 5 639 0.0 100.0 6 71 0.0 100.0 7 12 0.0 100.0 8 5 0.0 100.0 9 7 0.0 100.0 10 1 0.0 100.0 11 4 0.0 100.0 12 1 0.0 100.0 16 1 0.0 100.0 17 1 0.0 100.0 20 1 0.0 100.0 workman has a slightly better finger run length, with 92.8% of successive keys typed with different fingers, as opposed to mine at 91.%. Nice! > Also, do you not feel curious to try your own optimized layout? ;) LOL - I do, but I would be frustrated by the initial (and perhaps length) period of adjustment. Maybe I'm a layout creator, not a user! :o ### ### ### Whoops! Might need to make a little change to what you added to your website, I'm not the one who developed the Workman layout, I was merely referring to it and don't want to take credit for work that isn't my own. Sorry about any confusion.