Dvorak Keyboard Proved Better With Evolutionary Algorithm

The Dvorak keyboard recently came under some (relatively) high-profile fire from a 1996 Reason print feature that resurfaced on social news sites.

In 1932, Dr. August Dvorak patented a keyboard layout designed to reduce typist fatigue by placing keys to maximize hand alternation and minimize finger travel. It shouldn’t be too hard to improve on a 140-year-old keyboard that marketing people changed at the last minute so “typewriter” could be typed from one row. I’m sure you’re familiar with QWERTY.

Stan Liebowitz and Stephen E. Margolis argue that the studies supporting Dvorak are too biased to be reliable, specifically that Dvorak presided over some of the most-cited Navy tests. I’ve read claims to the contrary, but I came across something today that, in my mind, renders all of that moot.

Peter Clausler, a programmer with Cray (the most famous name in supercomputers), writes:

But Dvorak designed his layout in the 1930’s without the aid of computers. It contains a couple annoying features that lead to common errors in my typing — namely the placement of Y and B). Could a modern evolutionary algorithm and a huge input sample discover a better arrangement?

He took a measure of word frequency from 20 megabytes of the Project Gutenberg books, 10 years of email, and 100,000 lines of C code. Then he ran 4096 keyboard layouts through an evolutionary framework and allowed them to mutate, finally pitting the fastest against each other in an “all-star round.”

His initial results were disappointing. After a few days of use, it was obvious that the best layout was much less usable than Dvorak. He updated the algorithm to reward hand alternation and reflect the value of parallelism.

The resulting optimal keyboard looks an awful lot like Dvorak.

' , . p y  f g c r l   Dvorak layout
a o e u i  d h t n s   32129548
; q j k x  b m w v z

q w e r t  y u i o p   Sholes' layout, with quote replacing /
a s d f g  h j k l ;   59514344
z x c v b  n m , . '

k , u y p  w l m f c   Best evolved layout
o a e i d  r n t h s   28281895
q . ' ; z  x v g b j

He’s collecting data in support of another round of tests, but a very smart guy has designed a very smart system in which impartial machines find Dvorak to be much faster than QWERTY under a real-world ruleset.

But (perhaps not surprisingly), it sure looks a lot like Dvorak, too, and is not quantifiably all that much better, and that advantage is probably less than the level of error in my work estimation function.

I was confident of the benefits of Dvorak vs QWERTY within a few days of switching. Nice to see some empirical data.

Klausler’s experiments
Reason: Typing Errors