Re: model theory of error

Christian de Sainte Marie wrote:
>
> Michael Kifer wrote:
> 
> > How do you define an error independently of the evaluation strategy?
> > What does it mean to say that "RIF does not mandate any 
> > specific behaviour"? What is "behavior" exactly, if RIF (at least BLD) does
> > not define any evaluation strategy?
> 
> Let me try without using the words "error" or "behaviour"...
> 
> An evaluated function is defined over a domain, and it is undefined 
> outside of that domain.
> 
> If a function is used in a rule, we assume that any party that evaluates 
> that rule knows the domain of the function, whether it is specified 
> within RIF (builtin function) or not (application-specific).
> 
> So, anybody who may have to evaluate the function knows where it is 
> defined and where it is not, and is able to check, before evaluating it, 
> whether the arguments are in the domain, and the function defined, or not.
> 
> For the strict purpose of rule interchange, RIF needs to make sure that 
> all users have the same understanding of the rule - that is, draw the 
> same inferences - where the function is defined.
> 
> But does RIF need to guarantee anything beyond the common understanding 
> that the function is undefined, where it is undefined? Except, maybe, 
> that such cases must not be handled silently.
> 
> The same question applies wrt evaluated predicates.
> 
> Is that any clearer? And, if yes, does it make sense? And, if no, at 
> what step did I take the wrong turn?


It is clear like mud. You still fail to understand that we are supposed to
give formal semantics: model-theoretic, denotational, operational in that
order. We decided that for BLD we will give a model-theoretic semantics. If
you want to redefine the mission - fine.  But make sure you ask for another
12 months of extension.



	--michael  



> Cheers,
> 
> Christian
> 
>

Received on Thursday, 10 January 2008 01:18:38 UTC