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

[Nick Pelling <nickpelling@xxxxxxxxxxxxxxx>]

Hi everyone,

At 00:10 11/03/2005 -0600, Dennis wrote:
>         What effect does the size of the text corpus have?  The HMM 
> needed a text corpus of ~6 Mbyte for a clear result.  Would it tell us 
> something to create an enormous synthetic Voynichese corpus by Gabriel or 
> Jeff's method or Stolfi's Voynichese grammar and then analyze it?

Note that a grammar is a kind of generative probabilistic state machine, 
but not necessarily the same kind as described by Ben Preece.

For example, there's good reason to consider that EVA <o> functions 
differently in "qo" and "ol": while "qol" does occur in the VMs, it only 
normally occurs on pages where you typically find free-standing (ie 
non-"ol"/"al"-paired) "l" characters. So, a state machine where <o> maps to 
a single state would not be able to capture this behaviour satisfactorily.

In fact, this is basically true of any generative grammar where an 
individual letter (like <o>) appears in multiple "columns". The only easy 
way around it would be to pre-tokenise groups of letters (as Ben is already 
doing, though only for the usual suspects ATM), and compare those implicit 
transcriptions' n-state PFSMs.

>         What about defining a transcription alphabet that treats these 
> digraphs/verbose cipher elements as single glyphemes and then analyzing 
> the VMs in that transcription?    That seems like a useful exercise.

That's what I'm suggesting. :-)

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


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