When trying to type fast we often make mistakes. But these mistakes aren’t totally uncorrelated. Restrict to those mistakes where one has mistyped a letter for the correct one. For example, 1. `helo’ instead of `help’ (`o’ mistyped) 2. `systen’ instead of `system’ (`n’ mistyped). More often than not one would end up mistyping a letter which is proximate to the correct one on the keyboard.
If someone had time and energy, it would be interesting to collect large sample of mistakes from, say internet, and try to reproduce the keyboard modulo some group, say, reflection about the middle row. If one has the additional knowledge of the user being a right/left hander, one could keep this group size smaller.
