Engram publication and new approach and repository: https://github.com/binarybottle/engram

[binarybottle]

See https://github.com/binarybottle/engram for the new Engram approach and layouts!

After years of work, my open-access paper has just been published in the peer-reviewed International Journal of Human-Computer Interaction: "Optimizing comfortable keyboard layouts using human typing preferences and language-dependent n-grams: the Engram Study" (http://dx.doi.org/10.1080/10447318.2026.2665409) describing a new approach to optimizing keyboard layouts in different languages. Instead of assuming speed equals comfort (the paper shows speed explains only ~5.7% of preference variance, a poor proxy), I crowdsourced typing preference data from >500 people and used it to empirically derive ergonomics scoring criteria such as row separation, finger sequence, lateral stretch, and key preferences. I also validated which of Dvorak's original 1936 principles actually correlate with typing speed (spoiler: only 4 of 7 do).

These preferences drove multi-objective optimization over English and Spanish n-gram frequencies to produce new Engram-en and Engram-es layouts (https://engram-layouts.xyz/). All data, code, and layouts are fully open source. Please check out the article (and the Supplementary Material for my Engram Halloween costume!). Happy to answer questions.

Cheers,
Arno
arnoklein.info

[FishyLeaf57]

Interesting, nice work Arno. I do enjoy your layout, as you know I have been on it for years... Although sometimes I feel like I need to look to optimise it and or I often think have I chosen well? Anyway, like I said to you the other day, its been hard to learn the layout and even though its been years and many thousands of typing tests and training I sill have not reached my goals... So looking at yet another layout is not going to happen ... I very much enjoy engram v2 but I am struggling to get proficient, even though its all I use... lol... Anyway good work mate, all the best..

[binarybottle]

Thank you! Overcoming muscle memory is a real challenge. I found that I am able to learn a new layout faster by using two different physical keyboards -- one for the old layout vs. another for the new layout.

[rejuvenile]

I don’t see a BEAKL citation in your paper. Considering that predates your work significantly and is basically the same idea, that’s pretty poor scholarship.

[binarybottle]

Goodness, that's quite an accusation. BEAKL and Engram were generated using very different approaches. Engram uses multiobjective optimization based on crowdsourced typing preference data.

I included citations for published articles for layouts that I felt were historically relevant to my paper, and could not find a published article related to BEAKL. I remember looking on the website but couldn't find a published reference (http://ieants.cc/beakl/refs.php is blank).

In Table 1 I list all layouts that I evaluate against, including "BEAKL43 | 2023 | http://ieants.cc/beakl/".

[olddaniel]

That's great! Thanks for all the research and for your contributions to the community!

I've been using my version of Engram (English mixed with some Portuguese) for 2 years now, and I'm finally comfortable with it. Inevitably, when you learn that things can be (further) adjusted, it's hard to be truly satisfied with a version.

What attracted me to Engram in the first place was the data-driven approach. So I very much welcome updates backed up by more data! I'm excited to go down the modify-test-learn-try-again loop again!

Can you remind me if there's a more straightforward way to apply the Engram optimizer to 2 languages at once (with different weights, maybe?) and to a 36-key split keyboard?

Thanks again!

[binarybottle]

Thank you, @olddaniel -- I appreciate it.

The Engram approach requires 1-gram, 2-gram, and optionally 3-gram frequencies for a set of symbols. That's it. So if you decide how you'd care to weight the frequencies for two or more languages, then you should be able to replace the n-gram frequency files and run the same analysis as I did in my code. I haven't tried this myself, but in principle it should work just fine. Note that I focus on the 24 finger-column keys, so you have plenty more keys to work with!