A new proposal for the Hebrew keyboard

by Hadar Bar-David

The arrangement of the letters on a Hebrew keyboard was determined more than a century ago according to the common letters in Yiddish. In the 1980s, IBM copied the arrangement of letters to the keyboard, and from then until today this arrangement has been preserved.

In 2016, the Standards Institute tried to switch to a slightly different arrangement, which would avoid common mistakes and preserve the position of the punctuation marks in the transition between Hebrew and English, but the change was not successful.

To improve typing speed, I thought of an arrangement that would be determined according to the statistical chance of the appearance of letter pairs, or on other words: a Bigram. I discovered that LLM are based on bigrams!

I found a suitable database made by Haifa university.
With the help of Chat GPT I've organized the data in Google Sheet, and exported it as CSV format. I excluded final letters to simlify the experiment.

I loaded the data into this Colab. At first try the output made no sense (the letters remained in the same order).
I assumed it was because of rounding small numbers, so I made a new data set where all the numbers were multiplied by 10 million.
I ran it again and... It worked! This is the result I got:

To simulate the arrangement on a keyboard, I placed the proposal calculated by the Colab like a snake over three lines:

Ta-Da!

For the sake of the experiment, I loaded the same dataset into Chat GPT and requested:
"Can you create an optimizes order of the hebrew letters using the Travelling Salesman Problem?"
Chat GPT encountered a divide by zero problem, and resolved it by adding a small constant to the frequencies that are equal to zero.
This produced a completely different result:

I have no tools to measure which arrangement is better, in any case I was pleased to find my name there!