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

[Nick Pelling <nickpelling@xxxxxxxxxxxxxxx>]

Hi Ben,

At 16:17 09/03/2005 -0600, Ben Preece wrote:
> I'd hoped that this would be true, but I never seemed to see it in real
> languages (I mostly tried Latin), only in small toy languages that I
> invented just for testing.

Really, I think that Voynichese is only superficially language-like, and 
that it has an innate artificiality at its core. The idea behind examining 
its structure by incrementing the number of target states is to try to find 
a kind of transition point between the superficial language-like structure 
and whatever lies beneath.

>   Real languages apparently can't be
> realistically modeled without very, very large FSMs, so if a drop like
> this occurred, it may not happen until the FSM has hundreds of states.
> The largest I've ever tried to generate was 120 states.

Lucky then that Voynichese probably isn't a language (IMHO, of course). :-)

FWIW, I suspect the transition point will fall between 10 and 20 states 
(for a pre-paired transcription set, such as one where qo / ol / or / al / 
ar / iiii / iii / ii / eeee / eee / ee / dy / cfh / ckh / cph / cth / ch / 
sh / eo / od all map to single tokens). For EVA, the PFSM finder would also 
have to unpick all the tricky pairing rules correctly (lots of local minima 
to avoid), so I guess 60+ states would be closer there.

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


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