Re: VMs: VMs IS NOT Welsh

[Nick Pelling <incoming@xxxxxxxxxxxxxxxxx>]

Hi everyone,

At 13:59 24/01/03 +0100, Bernd Neuner wrote:
> The results of the paper recently mentioned by Rene (Akinori Ito, 
> Observation of left and
> right entropy in Voynich MS) clearly exclude that possibility (IMHO). 
> Though my knowledge
> of Welsh is rather shallow, I would expect that every Welsh word would - 
> as in other
> European languages - influence its immediate surroundings, that is, 
> increase/decrease
> the probability of being preceded/followed by certain other words.
>
> Example: Verbs tend to be follwed rather by non verbs.
>
> The Ito paper points out that this is not the case in the VMS. In other 
> words: the
> left/right word entropies suggest that the VMS symbol groups are virtually
> context-independent.
>
> Am I missing something?

Some quick thoughts on the interface between cryptography and statistics:-

Based on the common obfuscation trick used by the (very few) known 
cipherbets from 1440-1460 which have the same "4" character as the VMS, I'd 
say this: if the author of the VMS' cipherbet was the same person, there is 
a high (I'd say more than 90%) chance that the VMS also uses this exact 
same trick.

That is, as "4" and "4o" code for separate symbols in the other cipherbets 
where they appear, so too : similarly, ["o" + gallows] could well code for 
a separate symbol (whether a letter, index, or coding action) to plain 
[gallows]. This trick has two effects:-
(1) the size of the apparent alphabet is smaller than the underlying alphabet
(2) statistical analysis results become sharply less useful than on "pure" 
cipherbets

IMO, the problem with most letter-based statistical analyses done to date 
on the VMS is that they're merely superficially "rattling the bars of the 
cage" - that is, they're based on the (reasonable, but probably wrong) 
inference that, because of the clarity of the VMS's apparent alphabet, 
there's a one-to-one relation between the characters and the underlying 
alphabet.

I'm also highly suspicious of inferences based on "words" in the VMS, as I 
strongly suspect that not only do the apparent word "shapes" look a lot 
like a kind of embellished/obfuscated numbering system, but that the 
apparent length of VMS words also (as has been discussed many times before) 
has a non-"word"-like distribution.

Akinori Ito's paper would seem to support the (widely-held and long-held) 
view here that the code-stream' statistics don't look much like any 
existing language - but to the degree that it's still operating inside the 
cage built for us by the code's creator, it's not moving us forward.

In short: I think multiple letters frequently code for a single symbol, and 
that words are very frequently embellished numbers - so statistical 
analyses that don't try to break out of these twin traps *simultaneously* 
are very likely to be unhelpful.

But the question then arises: what kind of analysis *would* help?

Obviously, if it has any structure at all, then viewing it as a (probably 
complex) Markov chain might well be a decent starting point: but even so, 
this will require a great deal of insight to even get started - to date, 
I've had no great luck trying to model the numbering system used (which I 
believe is some kind of embellished Roman numerals), but perhaps some kind 
of state machine analysis might help move this forward.

Here's a suggestion: perhaps using Markov model analysis to find 
transitions in numbering patterns that *never* occur, and to then build up 
some kind of constraint analysis from that?

Conventional statistical tools wouldn't be much good at that, though. :-(

Cheers, .....Nick Pelling.....

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