Layout choices

[Steve]

Nice layout - I very much like the focus on ease of transition from Qwerty in bite-size steps.

I have some queries and observations I'd like to share:

I am curious about some of the layout decisions and wondered on what the basis they were made. It would be most interesting if the cumulative benefit from each change was shown. On the minimak homepage there is a table showing gains relative to dvorak for 4- and 12-key versions, but what about others, especially the 8-key version as that's the recommended one!

The 6-key minimak file I downloaded seems to switch R and F, even those these are not switched in the 8-key version. Is that a mistake? I would expect N-J switch is a much higher priority than either R-F or L-O?

The T key moves to Qwerty D position rather than F.  I would imagine swapping T with F brings a bigger benefit more quickly, as D is more common than F. Additionally, T would then remain on the same finger as Qwerty which would surely make it easeir to learn, and would have the side benefit of being consistent with other alternative layouts like Colemak, Workman, Norman, etc. Just a thought!

[Ted Lilley]

Thanks!  Those are good questions.  I spent a part of the weekend trying to do them justice, but couldn't finish.  I'll post a more thorough response in the next few days.

[Ted Lilley]

On Saturday, June 21, 2014 1:01:50 PM UTC-4, Steve wrote:

Nice layout - I very much like the focus on ease of transition from Qwerty in bite-size steps.

I have some queries and observations I'd like to share:

I am curious about some of the layout decisions and wondered on what the basis they were made. It would be most interesting if the cumulative benefit from each change was shown. On the minimak homepage there is a table showing gains relative to dvorak for 4- and 12-key versions, but what about others, especially the 8-key version as that's the recommended one!

 Good point.  The web page is overdue for a refresh.  When I put it up, my initial recommendation was the 12-key layout so that was the one highlighted in the original page.

After a good amount of time using the 12-key layout, I felt that the 8-key layout was a better fit for the goals of Minimak. I frequently switched between it and qwerty, and even after a good amount of time my speed on both were impacted more than I felt they should be. The 8-key layout is easier on the switch between qwerty, and I feel best recommending the layout which gives people a good chance to maintain the performance they're accustomed to while still providing strong benefit.

The 6-key minimak file I downloaded seems to switch R and F, even those these are not switched in the 8-key version. Is that a mistake? I would expect N-J switch is a much higher priority than either R-F or L-O?

 The 6-key minimak is a version that's meant for people who wanted to go to the 12-key version in two steps instead of three. For reasons I can't remember, that particular layout is better for that purpose.

The T key moves to Qwerty D position rather than F.  I would imagine swapping T with F brings a bigger benefit more quickly, as D is more common than F. Additionally, T would then remain on the same finger as Qwerty which would surely make it easeir to learn, and would have the side benefit of being consistent with other alternative layouts like Colemak, Workman, Norman, etc. Just a thought!

 One simple and reasonable answer is Ctrl-F. One of Minimak's stated goals is to not move common control key combinations if possible. That list includes:

Cut/Copy/Paste (X/C/V)

Select all (A)

Save (S)

Find (F)

Ctrl-D is the only left-handed home-position key combination which doesn't have a strong use like A, S and F (even though I happen to use it all the time to close Linux shells since it's the Unix end-of-file character).

That's a good reason, but it's not the primary driver though. The real reason is that the D position scores better on the keyboard layout analyzer (http://patorjk.com/keyboard-layout-analyzer/) which is the metric on which I rely. The F position scores well also, just not as well. When you look more closely at the reasons why the D position scores better, I find them convincing.

If you put T in the F position, the one metric which scores better than the D position is reduction in finger movement distance (3.4%). That's an important metric, however you get worse balance of finger usage (15%) and worse same-finger repetition (3.9%). I find the latter two outweigh the former and the control-key argument seals the deal.

By way of explanation, finger movement distance pretty intuitive. It's the amount of moving around your fingers do on the keyboard and presumably the less you do of it, the better. The typical way to achieve less distance is to bring the most frequently-used letters, such as T, into the typing home position (ASDF JKL;) where your fingers rest and don't have to move at all to strike. So putting T in the F position is a win, it's just not a huge win over the D position since it's only a 3% reduction by comparison.

You can see the graph on distance here: https://drive.google.com/file/d/0ByCaT94mpmKwVmphRWg1dzYzT3M/edit?usp=sharing

The comparison is between two different version of the Minimak four-key layout. E and K are swapped, so we can see the relative performance of the two T swaps, D and F, as they would perform in the Minimak setting. This is important because the relative performance depends on not only how those keys are situated but also how they relate to pretty much every other key on the layout.

The graph can be a bit tricky to read at first. It lays out the fingers of both hands on the x-axis. Above each finger are the scores for the both layouts. The darker one always on the left is when T is in the D position. The right is it in the F position.

As you can see, every other finger but the left middle is identical in movement distance, as you'd expect. The left middle finger is better (shorter) with T in the F position, the one on the right.

Now look at the graph for finger usage: https://drive.google.com/file/d/0ByCaT94mpmKwT2ZLNjVRV1RYcFE/edit?usp=sharing

Finger usage is simply the number of times you used the finger while you type. It's like finger distance, except that distance doesn't include when you type a key in the home position since that's "at rest", while usage does. Each key strike counts the same for usage (1), while further keys add more distance. So they're related metrics, but they aren't the same and they both have useful information.

The desirable trait for finger usage isn't reduction, it's balance. You want to spread the load between your fingers rather than over- or under-utilize them. While the ideal distribution is debatable, I generally think it's useful to spread usage as evenly as you can (with an exception for the pinky).

In the picture you see that again, the two layouts are mostly identical. The difference is in the middle and index fingers of the left hand this time. Putting T in the F position (the lighter column over each finger) puts very high usage on the index finger. It's almost as much as the high mark of the right hand. The right hand is a problem as well, just one we can't address in a four-key layout such as this. In addition, the middle finger gets very little usage, tied with the other least-used, the left ring finger. It's arguable, but I'd call these over- and under-utilized respectively.

In contrast, the D position puts most of the usage on the left middle finger, and less on the index. However, the columns are closer in height to each other, which makes this more balanced. If you want to get numeric about it, you can compare the standard deviation of finger usage in each layout and see that the D position reduces deviation (is more balanced) by about 15% compared to the F position.

Finally, take a look at the graph of same-finger repetition: https://drive.google.com/file/d/0ByCaT94mpmKwSGNMMDVTZlpvMkU/edit?usp=sharing

While finger distance, balance and other metrics are important, I have a particular distaste for same-finger repetition. You don't notice it if you aren't looking for it, but once you're aware of it, it becomes an annoyance. Same finger repetition is when you have to use the same finger twice in a row (for different keys), which slows you down because you can't be positioning a different finger to strike the following key while you are in the process of striking the first key. While it doesn't overshadow the other metrics, I put more weight on it than, say, movement distance.

Ignore the massive right-handed pinky column in the graph, it's skewed by it's responsibility for all of the punctuation and special keys on the right side of the keyboard, which the keyboard layout analyzer dutifully includes.

The rest of the graph maxes out (max being bad) on the right-hand index finger. That makes sense because it's responsible for six keys, second only to that pinky. Most are only responsible for three keys.

If you assign T to the F position (light blue), it's no worse than QWERTY since all you've done is change two keys on the same finger. Still, it shares the title of worst (not counting the pinky) with the right index.

If you put it in the D position however, the index repetition goes down quite a bit. Even if you add back in the increase for the middle finger, it is now less than both the right index as well as middle finger repetitions. That's a win. Overall repetition goes down 3.9%. Not earthshaking, but more than the decrease in finger distance (3.4%) already and as I said, I weight it more anyway.

Add to that the other arguments and that's what capped my decision.

I should mention that there are other important metrics I used in Minimak, such as hand alternation and balance of usage between hands. While you can't optimize for everything, it's interesting to look at them all and see how the layout you're using works. I'd recommend playing with the keyboard layout analyzer and really looking at all of the information it provides.

If you want to reproduce any of my tests, I used the full text of Alice in Wonderland, which seems to align well with the standard key frequencies for English which are referenced in many places. You can find it here: https://code.google.com/p/nltk/source/browse/trunk/nltk_data/packages/corpora/gutenberg.zip As a shortcut, you can just use the kla's built-in sample of the first chapter of Alice which it makes easy to load, but the full corpus is better.

As for comparing the Minimak design decisions with those of Colemak and company, I wouldn't try to draw a lot of inferences between different layouts. They can't be looked at in isolated pieces, they only make sense in the whole. The utility of every key literally depends on the placement of every other key on the keyboard. Not only that, designers pick out different goals. If one prefers minimizing repetitions like Colemak does, you'll get different design decisions than Workman makes.  Besides, if every other layout jumped off a bridge, should Minimak? ;)

Thanks for your question, it really did make me think and reinvestigate before I answered.  I even found that the true reasons weren't the first ones I thought they'd be, so I learned something in the process.

Ted

[Steve]

Thanks for your answer, a very detailed and informative explanation. Regarding the T location, I guess the killer reason is the reduced same-finger repitition. As you say, Colemak et al address this by changing more keys. Probably I prefer some degree of overloading on the index finger relative to others, as I consider it my strongest finger. But given you are only changing 4 keys at that point, the T/D switch makes sense.

I guess my question was partly brought about as I had considered trying out Minimak as a possible intermediate step, eventually reaching Colemak at a later point. Indeed, apart from the notably different T position, Minimak shares a lot of similarities with Tarmak.

Thanks for the explanation on the 6-key version. Personally I think two separate tracks (4-8-12 vs 6-12) could add confusion and it would be better if there was only one Minimak "track" to go down. I'm guessing in that case the next change after Minimak-4 would be the N/J switch. If I had decided on Minimak(*) and wanted the optimal balance of benefit vs changes, I'd probably pick that.

(*) In the end I decided to go with Colemak (currently about 6-weeks in), but I still think the minimal-change concept is good, and I hope it gains more traction!