Re: VMs: Proof, a la Rugg. that the german language is nothing but a hoax (shorter and perhaps more interesting)

[Gabriel Landini <G.Landini@xxxxxxxxxx>]

On Friday 10 September 2004 19:55, Michael Winkelmann wrote:
> The "nonsense" text generation can be done by using a (well created)
> three-dimensional array of character frequencies and a random number
> generator in a very simple algorithm. 

This is what you can also do with Monkey: n-transition probabilities. This is 
described in chapter 4 ("Language") of Bennett's book (which also deals with 
the VMS and also in an old "Computer Recreations" or "Mathematical 
Recreations" (I forgot which) of Scientific American (ha!). 
There is also a *very* old DOS programme call "Mark V Shaney" (for Markov 
chain) that reads a text input and outputs gibberish of the desired order 
(which is described in the Sci Am column).
When you use words instead of characters and order 3 or 4 for the transitions 
then the text "seems" to make some sort of sense and it can be funny (or sort 
of).
This does not prove anything, it just generates text with similar statistics 
to the input text which span *up to that order used*. Larger scale structures 
are non-existent.

> If someone does not understand the 
> german language and creates fragments of gibberish text by populating an
> array and applying an algorithm to that data, which creates "german
> looking" text with some statistical properties near to real german
> texts, what does it proof; what a kind of evidence is it? 

It is evidence of the "no evidence" kind.

> For Rugg it 
> seems to proof that german is gibberish - and the "problem of the german
> language" is solved.

Some things that Rugg's solution cannot explain:

It does not explain the statistics of the labels. The argument that then one 
generates a further table only for lables is, to me, naive.

Weirdoes. If one makes N cuttings (I think Rugg suggests 3) and slides the 
card thought the table, then whenever one hits a weirdo in the first slot, 
the same character should appear later in the other positions. More precisely 
the weirdo should appear N times when using a particular table if there is a 
single cell containing it. This does not seem to happen as most weirdoes 
appear typically only once in the vms. This seems incompatible with the 
concept unless the weirdoes were located in single cells at the edges of the 
table (where the sliding card cannot find the other offset characters). I 
could accept this for a few occurrences, but not for all of them. Of course 
one can also conveniently claim that the table may not have been used 
strictly in sequence or something like that. But having to resort to this 
kind of explanations makes the whole argument even weaker.

I also think that analysing Cardan gibberish it may be possible to reconstruct 
the width of the table and the relative position of the cuttings based on the 
distances between the appearances of weirdo characters that appear exactly N 
times (or even multiples of N). Of course, one does not know N, but perhaps a 
brute force approach would do.
if no weirdoes appear, then one needs a lot further explaining to do because 
they do appear in the vms.

Cheers,

Gabriel (still unconvinced)

______________________________________________________________________
To unsubscribe, send mail to majordomo@xxxxxxxxxxx with a body saying:
unsubscribe vms-list