5 ms·

You don't need a keylogger. Just take some things you've written and count the characters. It won't be exact since it excludes rewrites but the distribution sho

by jryb 2y ago

You don't need a keylogger. Just take some things you've written and count the characters. It won't be exact since it excludes rewrites but the distribution should be very close.

serial_dev 2y ago
The SwiftKey keyboard (mobile) used to (maybe even today) ask for your Gmail account, goes through your stuff, recognizes your style and includes your vocabulary. Ignoring the obvious privacy issues, it provides great suggestions and learns some words that are not in the other keyboards' dictionaries. Maybe you could apply the same for aiding an algorithm to create a personalized keyboard layout.
yjftsjthsd-h 2y ago
I'd like to also include keyboard shortcuts and commands that don't end up in recorded text.
- jryb 2y ago
  At least for any defined set of layouts, you could infer what those other keypresses would be and compare their efficiencies. Still a compromise but it satisfies all the security issues.
taeric 2y ago
If you do go down this rabbit hole, you probably want to optimize hand alternation. That is, it is not enough to keep the high value letters home row, you probably want to split them between hands, while keeping them home row. Similarly, the slowest combinations of characters will be the ones that force a finger to go from top to bottom row. On qwerty, as an example, 'cemetery' would be pretty bad for the opening two letters.
eviks 2y ago
This bad approach is why backspace and cursors keys are ignored in many layouts because they don't appear in the text corpus, which is a bigger issue than rewrites You do need a keylogger, and a context-aware one at that for proper analysis, otherwise it's better to just use an existing layout, won't get much extra from personal corpus anyway, so not worth the effort to do a poor personalized analysis

5 ms·

You don't need a keylogger. Just take some things you've written and count the characters. It won't be exact since it excludes rewrites but the distribution sho

by jryb 2y ago

You don't need a keylogger. Just take some things you've written and count the characters. It won't be exact since it excludes rewrites but the distribution should be very close.

serial_dev 2y ago
The SwiftKey keyboard (mobile) used to (maybe even today) ask for your Gmail account, goes through your stuff, recognizes your style and includes your vocabulary. Ignoring the obvious privacy issues, it provides great suggestions and learns some words that are not in the other keyboards' dictionaries. Maybe you could apply the same for aiding an algorithm to create a personalized keyboard layout.
yjftsjthsd-h 2y ago
I'd like to also include keyboard shortcuts and commands that don't end up in recorded text.
- jryb 2y ago
  At least for any defined set of layouts, you could infer what those other keypresses would be and compare their efficiencies. Still a compromise but it satisfies all the security issues.
taeric 2y ago
If you do go down this rabbit hole, you probably want to optimize hand alternation. That is, it is not enough to keep the high value letters home row, you probably want to split them between hands, while keeping them home row. Similarly, the slowest combinations of characters will be the ones that force a finger to go from top to bottom row. On qwerty, as an example, 'cemetery' would be pretty bad for the opening two letters.
eviks 2y ago
This bad approach is why backspace and cursors keys are ignored in many layouts because they don't appear in the text corpus, which is a bigger issue than rewrites You do need a keylogger, and a context-aware one at that for proper analysis, otherwise it's better to just use an existing layout, won't get much extra from personal corpus anyway, so not worth the effort to do a poor personalized analysis