Human random number generators

["Gabriel Landini" <G.Landini@xxxxxxxxxx>]

Hi all,
The other day I promised a reference. The paper is by Schenkel, 
Zhang & Zhang (1993) Long range correlations in human writings. 
Fractals Vol. 1 No.1: 47-57.

The paper is not particularly good I think that there is a problem in 
the way of coding letters:
Each letter is arbitrarily coded as a binary number of 5 bits. The 
sequence of 0's and 1's is converted to a random walk by 
assigning increments of  +1 or -1 depending whether the current bit 
is 1 or 0.
My criticism is that if one assigned "11111" to a common letter then 
there will be a bias on the random walk, but this would not show up 
as strong if the same letter was coded (let's say) as "10101".
The authors argue that those correlations would not show up 
beyond the word level. I am not sure that it would be the case and 
they do not show any evidence for that either.

They show also long correlations in a human-written sequence of 
"random" numbers.

Their conclusions are fairly inconclusive.

Cheers,

Gabriel