• You are not logged in.

    Keyboard Effort Grids

    • Started by stevep99
    • 8 Replies:
    • Reputation: 88
    • From: UK
    • Registered: 14-Apr-2014
    • Posts: 837

    I have been working recently on creating a new model to generate keyboard effort grids.

    The idea behind this is to reduce the amount of guesswork and personal preference that is involved - including in my own - and increase the amount of objective maths! To this end, I have created what I hope is a more accurate - or at least a more systematic - model of keyboard effort!

    You can read about it here

    My newly generated effort grid is at the bottom of the page if you want to skip over the details.

    Last edited by stevep99 (21-Feb-2021 12:49:03)

    Using Colemak Mod-DH with some additional ergonomic keyboard mods.

    Offline
    • 0
    • Reputation: 18
    • From: CZ
    • Registered: 14-Feb-2012
    • Posts: 440

    Garbage in, garbage out.

    Not all fingers are created equal, and there is broad agreement that the index and middle fingers are strongest, the pinkies the weakest. Quantifying the relative strengths of each finger is a subjective matter, but in this model, the effort to type a given key is multiplied by a factor based on the finger used: [some RNG output]

    Alternatively, you could at least look up studies on finger strength or something... Assuming that finger strength is actually *the* important factor.

    In common with my fork of the patorjk analyzer, this model also assumes lateral motion of the hand is more costly than simply curling a finger inward or outward. Consequently, motions that are transverse to the direction of the forearm are penalized more heavily than directions that are aligned (the mesial direction).

    Wait, but why?

    I'm increasingly convinced that these models have very little to connection to the way people [do/should] type. The fundamental observation is that typing is not a sequence of isolated keystrokes, but a complex dynamic process that is better represented by hand positions.

    [assume QWERTY labels] When a typist writes 'rt', why should they return their index finger to the home row between the two keystroke? why should they use only the index finger and not a pair of adjacent fingers? When a typist writes 'in' or 'on', why should they perform uncomfortably looking row jumps, instead of rotating their whole hand (forearm) a bit?

    Sigh. It would have been easier to simply copy jacobolus' post:

    All the programmatic layout analyzers I’ve seen are based on totally arbitrary heuristics, made up by their authors with little or even no support in any kind of evidence, or even convincing reasoned argument.

    What’s needed to make a reasonably scientific model is to recruit a number of very experienced and effective typists, ideally across a few different logical keyboard layouts, and record the timing of their keystrokes when handling different combinations (ideally not just trigrams, but groups of 5–6 letters at a time) of letters with high precision, and then do some thoughtful and comprehensive data analysis, trying to figure out which letter combinations are fast or slow, accurate or inaccurate, and how the speed and accuracy relates to other factors like the break-down of a text into sentences, words, syllables, and common character combinations. Ideally high speed video would be captured, so that particular notable patterns in the data could have the associated hand movement patterns directly observed.

    The reason this is necessary is because different typists have different fingerings for particular keys (or even varying fingering depending on context) and I suspect different ways of chunking up sentences and words into finger movements. By looking at the way typists who use different logical layouts deal with the same text, it should be possible to learn the way that word spelling chunking interacts with typing motion chunking, and figure out how to best align the two on a keyboard layout. Furthermore, it should be possible to learn (with evidentiary support) which finger motion combinations are most natural or most awkward for experienced typists at high speed, and to analyze the differences between different typists who use different fingering patterns, hold their hands differently, have different hand size/shape, etc.

    But the existing layout analyzers have models such as measuring the physical distance in millimeters from the “home row” key for each finger to each other key, and then using that distance as a proxy for typing difficulty, arbitrarily assigning different fingers a score and then scoring keys based on which finger is used, arbitrarily assigning different rows on the keyboard a score, giving an arbitrary score to two-letter combinations on different hands vs. the same hand, giving a score based on the percentage of letters typed with one hand or the other, etc. All of these measures are complete crap, and should not be used for any kind of serious purpose. They have no typing data to back them up, but are purely conjectural.

    The carpalx method, to take one example, is a total joke. It has no reasonable basis for any of its scoring model, but it masks that by making the model so complex to understand that it’s impossible to guess what the relative scores will be for two letter combinations without explicitly working out the details. (Then if you actually compare specific combinations of letters and do work out the details, the scoring bears no relation whatsoever to real-world experience typing the same combinations on a physical keyboard.)

    Get yourself an ergonomic keyboard. Learn Maltron. Or stenography.

    Offline
    • 0
    • Reputation: 2
    • Registered: 04-Aug-2019
    • Posts: 9

    Looks interesting Steve. I will have a read through this at the weekend. Thanks for sharing!

    Offline
    • 1
    • Reputation: 149
    • From: Oslo, Norway
    • Registered: 13-Dec-2006
    • Posts: 5,007

    It's a very interesting effort, but your results so far puzzle me. I fear that much insight remains to be found?

    keyb_calc_iso_scores.png

    Do I interpret your figure right, in that a weight of 2 is considered "twice as hard" as a weight of 1, etc? If so, that figure doesn't scan at all. Hitting Colemak F is nowhere near twice as hard as hitting S, and hitting the right Enter key certainly isn't 12 times as hard as S/T/N/E! Come on, which would you rather do – hit Enter once or type TN six times (or STNE three times to avoid same-finger fatigue, whatever)? This doesn't scan. Please enlighten me.

    Using the Curl home position, my middle finger rests at the top of the key. The distance to the key above is very low. I suspect that the distance varies between typists depending on habits, technique and finger lengths. I'm really not sure how to reflect those factors.

    Last edited by DreymaR (22-Feb-2021 14:55:00)

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

    Offline
    • 0
    • Reputation: 88
    • From: UK
    • Registered: 14-Apr-2014
    • Posts: 837
    davkol said:

    Garbage in, garbage out.

    I don't think your aggressive and dismissive tone is warranted, but anyway, to address your points:

    davkol said:

    Assuming that finger strength is actually *the* important factor.

    The model doesn't claim the finger strength is *the* important factor, it includes it as one factor amongst others. Yes, the assignment of relative strengths is subjective, a point I acknowledged, but I don't think it's controversial to say index fingers are strong, pinkies are weak, so the question at issue is to what extent and how it should be quantified. I also wish there were more good studies of keyboard use. If you are aware of a study that could better inform how to treat relative finger strengths, please post a link to it instead, at least that would be constructive...

    davkol said:

    Consequently, motions that are transverse to the direction of the forearm are penalized more heavily

    Wait, but why?

    I'm increasingly convinced that these models have very little to connection to the way people [do/should] type. The fundamental observation is that typing is not a sequence of isolated keystrokes, but a complex dynamic process that is better represented by hand positions.

    The lateral hand motion being more costly is not novel, but increasingly mainstream - the Colemak-DH and Workman layouts are both based around it, and there are plenty of people around who agree with this premise, so for those people including this is perfectly valid. I agree with your other point about sequences of keystrokes being significant, this keyboard effort grid is one part of a complete model, something I did mention on that page.

    davkol said:

    [assume QWERTY labels] When a typist writes 'rt', why should they return their index finger to the home row between the two keystroke?

    Again the model isn't saying that. Yes, it's assigning a value for each key representing the how easy it is to reach on average from the home position, which I'd argue is a useful piece of information to include.

    davkol said:

    (posts long rant)

    For some reason you seem to think that unless people are able to recruit a team of typists and fund a huge study, that they shouldn't be allowed to think about and attempt to quantify what factors they think are relevant to typing efficiency and comfort based on a combination of their own experience, discussions with others, and the application of mathematics. I've critiqued many of the existing models that are out there previously, including the carpalx, workman and beakl ones - indeed, this piece of work is attempting to address some of what I see as some of the flaws and relevant but missing factors in other models.

    Yes, it's just a model, no-one is claiming it's perfect. As with any model, the extent to which you accept its conclusions depends on how much you agree with its premises. People are free to agree or disagree with elements of the model, or suggest changes - it is still a WIP and I'm open to suggestions, hence my posting it here. No need to be triggered by it.

    Last edited by stevep99 (22-Feb-2021 18:35:44)

    Using Colemak Mod-DH with some additional ergonomic keyboard mods.

    Offline
    • 0
    • Reputation: 88
    • From: UK
    • Registered: 14-Apr-2014
    • Posts: 837
    DreymaR said:

    I interpret your figure right, in that a weight of 2 is considered "twice as hard" as a weight of 1, etc?

    I wouldn't interpret it quite as literally as that, from a keyboard layout algorithm perspective what matters is that the relative values of the different keys rather than some absolutely strict measure of difficulty. Remember, we are coming from a situation where people tend to score the best home keys at some base value (usually 1) and then the scaling factor is somewhat arbitrary after that:

    keyboard_scores_ergonomic.png

    400px-Keyboard_effort_grid.png

    keyboard_graded1.png

    czKUnv5.png

    Perhaps that's why people like davkol are so against doing this kind of this thing, but I still think there is some value to be had if the parameters of the model have a reasonable basis, and are adjustable to reflect personal preference and experience.

    DreymaR said:

    Hitting the right Enter key certainly isn't 12 times as hard!

    Yeah, I'm not happy with that either, it does seem excessive. Nonetheless, it's fair to say enter should be penalized heavily, as you entire hand loses the home position, which massively ruins your flow. Especially the ISO enter without the wide mod, which is what we are talking about here.  But I'd agree the transverse penalty and pinky penalties are combining here in a unfair way, so I will certainly look to tweak the algorithm there. Update: Yes, there was effectively a double-penalty for pinky keys far out on the sides. I have addressed this, the results are more reasonable - but I also think the model should mainly be used to consider keys closer to the home position. There comes a point where keys are so far away they are not useful practically for comfortable, flowing typing.

    In the most part the result has come out similar to my original (guesswork) values on the DH analyzer page, which gave me confidence that it wasn't miles out, but obviously the analyzer is designed around the 30 main keys only.

    DreymaR said:

    I suspect that the distance varies between typists depending on habits, technique and finger lengths. I'm really not sure how to reflect those factors.

    This is an absolutely valid point. By moving your entire default hand position higher or lower, you can affect the relative scoring of the top vs bottom rows. If you were using a layout that heavily emphasized the top row for example, you may well have your hands higher, and so the scores would reflect this. Personally my hands tend to be somewhat lower, my index fingers are on the edge of T and N, and the middles aren't really at the top of S and E.  Perhaps an adjustable parameter is needed which would model this hand positioning factor.

    Last edited by stevep99 (22-Feb-2021 19:07:57)

    Using Colemak Mod-DH with some additional ergonomic keyboard mods.

    Offline
    • 1
    • Reputation: 149
    • From: Oslo, Norway
    • Registered: 13-Dec-2006
    • Posts: 5,007

    The relative weighting of keys near and far away will also have a measure of individuality to it. When you see top QWERTY typists at it, they don't seem to mind having their fingers fly all over the board like crazy angry birds. When you talk to the endgame mech board people they seem to detest having more than 30 keys so it's a totally different game. For most of us, neither of those extremes will be the right answer.

    As mentioned your effort is very interesting but it has to be interpreted right of course. Davkol's GIGO comment is unwarranted, but its core of truth is that any modeling will be subjective and hard to interpret and do just right. I think this is a promising step forward from just plopping subjective numbers into a matrix: Even if such number-picking can take into account more complexity whether consciously or subconsciously, it's more opaque and highly subjective "mathemagicks".

    I'd still paste a disclaimer with rather large letters on your analysis as such, at least for now! It seems to me that an increasing number of people use your analyzer to create a layout with the highest possible score they can get, and then tell the world – or at least the Monkeytype Discord #layouts channel – that this is now the new best layout ever! You need to lower their expectations a bit, and make them aware of the challenges of modeling. At the very least, make them understand that a difference of less than, say, 1 point score may just as easily stem from a model bias or differences in corpus or keyboard geometry as from a real difference in layout quality! Probably a bit more than 1 points, actually, as Viper's layout analysis showed us.

    And again, Davkol isn't wrong when he points out that most of this is still conjectural and not backed in hard science. Which we don't have the resources to perform. Much of it makes sense intuitively, but solid proof is very hard to come by. Let's have fun and gain insights in whatever way we may, but never lose sight of the limitations to it.

    But for now: Again, kudos on an interesting effort.

    Last edited by DreymaR (23-Feb-2021 10:46:06)

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

    Offline
    • 0
    • Reputation: 88
    • From: UK
    • Registered: 14-Apr-2014
    • Posts: 837

    Fair points DreymaR. The truth it we are working from a low base of hard knowledge - modern typing practices and keyboard usage are not a fashionable things to spend research budgets on, and in any case most ordinary people are extremely resistant to even modest changes to their keyboard, which means there's little demand for such research anyway. So in the main it's left to self-selecting enthusiasts who have developed an interest. I wasn't aware my analyzer is popular on Monkeytype discords, but TBH there aren't actually many user-friendly analyzers out there to choose from so if so it's probably only popular by default! That said, the fact that it's transparent with the assumptions it makes and has adjustable parameters so that people can customize it reflect what they believe are important factors for them, does count in its favour.

    As we know, the effort grids that are out there are generally just arbitrary numbers invented by their creators, including the (default) values in my DH analyzer. I think having the values calculated rather than just declared is a step forward, as there is then an algorithm that people can debate the merits of / improve etc. That way people can evaluate the underlying mechanism that gave rise to the result, rather than the end result itself. It is already caveated with disclaimers I feel - including the important point that a keyboard effort grid is just one tool that can be used to evaluate layouts, and should not be thought of as a complete model of keyboard of itself.

    Last edited by stevep99 (23-Feb-2021 12:40:13)

    Using Colemak Mod-DH with some additional ergonomic keyboard mods.

    Offline
    • 1
    • Reputation: 149
    • From: Oslo, Norway
    • Registered: 13-Dec-2006
    • Posts: 5,007

    Surely you've seen the same effect at the Colemak Discord: Your analyzer is interpreted as ground truth. I just wish people would read the fine print more closely.

    I'm not sure exactly how common it is right now, but after first Nyfee and then NotGate based their layout evaluation on your analyzer it's had lots of traction it seems. Congratulations on your celebrity status, mate!  (✿◠‿◠)

    Last edited by DreymaR (23-Feb-2021 14:58:00)

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

    Offline
    • 0