Thread Links Date Links
Thread Prev Thread Next Thread Index Date Prev Date Next Date Index

Re: SUO: RE: A disclaimer about 'new KIF'




>John and Pat:
>
>This is just the sort of discussion which I should like to see more of, and
>which I consider essential to achieving clarity in a SUO.
>
>However, although I cannot tell with certainly, I think that you may be
>discussing at cross purposes, since John refers to some utilization of a
>hierarchy of meta-languages, while Pat speaks only of the expressiveness of
>'sorted FOL'.
>
>Clarification?

I will try.  Background: part of the motivation for a 'new KIF' was a 
perception that KIF, by trying to be aqll things to all mean, was 
rather a complex and awkward beast, and that a stripped-down version 
might be of greater utility. So the 'newKIF' is being designed to 
have a 'core' which is as simple as possible, and several 
'extensions' on top of the core which are kind of optional. The core 
is now pretty well understood, and it will be similar to KIF 3.0 but 
with a simpler syntax and a more powerful semantics, and will allow 
superficially "higher-order" operations like quantifying over 
relations and functions. However, it will not actually be 
higher-order. (It is equivalent to a sublanguage of Lw1w which I 
think is compact, but havnt quite proved this yet.) The extensions 
are somewhat murkier than the core, but two of them that have been 
fairly well identified are (1) a sorted extension, in which for 
example argument 'sorts' for functions and relations can be specified 
and will be checkable at parse time; and (2) a meta-extension, which 
will have fully-fledged capabilities for describing its own syntax 
(and the syntax of other languages).

Ive been working on the sorted extension but as far as I know nobody 
has been seriously working on the meta-extension.

The sorted extension provides no actual logical expressivity beyond 
the core; it only 'builds in' some of that expressivity into the 
syntax, which many people find convenient. The meta-extension will 
need a fully fledged mechanism for quasi-quotation and a 
truth-predicate, at the very least, and probably some extra inference 
rules for upward and downward reflection, so it will have a 
considerably extended semantic theory which has not yet been fully 
worked out. It ought to be capable of describing meta-languages to 
any depth,in principle, though I imagine that it would rapidly become 
unworkable in practice beyond two metalevels.

However, none of these will be genuinely higher-order. All extensions 
of the language will be compact and have finite axiomatisations, and 
will be efficiently implementable.

Hope this helps.

Pat Hayes

---------------------------------------------------------------------
IHMC					(850)434 8903   home
40 South Alcaniz St.			(850)202 4416   office
Pensacola,  FL 32501			(850)202 4440   fax
phayes@ai.uwf.edu 
http://www.coginst.uwf.edu/~phayes