RE: VMs: RE: RE: Best-fit 2-state PFSMs

[Rene Zandbergen <r_zandbergen@xxxxxxxxx>]

--- Ben Preece <ben.preece@xxxxxxxxxx> wrote:

> A PFSM of course is a pair of functions p() and x()
> on a set S of states and set T of tokens.  If s is a
> state and t is a token, then p(s,t) is the
> probability that t will be the input token when the
> PFSM is in state s, and x(s,t) is the next state
> that the PFSM transitions to when t is the input
> token in state s.
> 
> Suppose t_0, t_1, t_2, ... is the sequence of input
> tokens.  Generate the sequence s_0, s_1, s_2, ... of
> states, where s_0 is the initial state, and s_n+1 =
> x(s_n,t_n).  Then the total information generated by
> the input text is
> 
>     sum log2(1/p(s_n,t_n))
> 

Very interesting indeed. 
The actual value of the sum should give some 
kind of a 'score' of the separation, although it
should be treated carefully since it will depend
not only on the success of the separation, but also
on the size of the alphabet. Still, having seen that
vowel/consonant separation in Voynichese is
not as clear-cut as e.g. in Latin, I wonder if
your 'score' would say something about this (once
you've done it for Voynichese).

I strongly suggest not to do it in Eva, but to
use something similar to what I used on this
page:
  http://www.voynich.nu/extra/curabcd.html
(the column labelled Curva).

I have occasionally wondered if a three-state
model would be able to pick up the liquids/semi-
vowels (not too sure about the right terminology)
that build up such combinations as:
 -pr-  -pl-  -kw-  -fr- 

I also wonder if you had a look at:
  http://www.voynich.nu/extra/wordent.html
where a similar quantity to the one used by you
is used to look at the
information contained in the VMs text.
Sorry for the figures at the end. They are hard to
figure out since at the time I decided to put
too many graphs together in single plots.
I don't have the numbers anymore, otherwise I'd 
do them again.

Cheers, Rene
 


	
		
__________________________________ 
Celebrate Yahoo!'s 10th Birthday! 
Yahoo! Netrospective: 100 Moments of the Web 
http://birthday.yahoo.com/netrospective/
______________________________________________________________________
To unsubscribe, send mail to majordomo@xxxxxxxxxxx with a body saying:
unsubscribe vms-list