Re: VMs: Slightly modified VMS generator output

[Nick Pelling <incoming@xxxxxxxxxxxxxxxxx>]

Hi Jeff,

At 23:35 03/12/2003 +0000, Jeff wrote:
> Every one here is making the unfounded assumption that because a word that
> is generated by an algorithm does not appear in the VMS text it cannot be
> valid. In the case of an artificial language the VMS will be a subset of a
> universal set. If the method can generate every word in the subset then how
> can it be said that the extra words are not part of the universal set that
> the author chose not to select?

No, it's only you that's making that assumption. What we're *actually* 
trying to do is to generate models for generating text that are strongly 
predictive of the text as observed (ie minimising the size of the 
[negentropic] residual), whilst simultaneously trying to minimise the size 
of the model (ie minimising the number & complexity of assumptions).

AIUI, these two values are essentially what the Bayesian Information 
Criterion (BIC) tries to merge into a single composite measure.

IMO, what we have so far failed to do is to operationalise our ideas fully 
- what is the BIC of the OKOKO model, or of my four-column model? How much 
information do the three (quite reasonable) assumptions mentioned by John 
Grove discard? etc

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


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