Re: DAML and reification

From: pat hayes ([email protected])
Date: 03/05/01


>At 12:42 PM -0500 3/5/01, Peter F. Patel-Schneider wrote:
>>The problem is just what is supposed to be the meaning of reification in
>>RDF.  There is no guidance from the RDF spec in answering this question.
>>(Consider the DAML axiomatization---it provides no meaning for
>>reification beyond the totally uninterpreted data structures.)
>>
>>Should we be answering this question in advance of RDF answering it?  I
>>don't think so.  DAML+OIL is an ontology language, not a language for
>>representing and reasoning about statements.
>
>While I know what you mean, there are some people (probably 
>including me) who would see these clauses as contradictory.

Well, not including me.  Peter is absolutely right to insist on 
keeping these issues apart. Ontologies *describe* domains of one kind 
or another (mostly made of classes and properties and similar stuff), 
and *consist* of statements. Now, one possible kind of domain 
consists of (classes of) statements, of course, but:
1. those are very idiosyncratic domains with their own very 
idiosyncratic properties, and not at all typical of domains in 
general, so may well require special treatment;
2. if the domain being described is supposed contain the very 
statements that are being used to talk about it with, then Peter's 
nuclear-power metaphor is particularly apt, as there is then the very 
real possibility of semantic melt-down. Reading the bright enthusiasm 
with which RDF-ers point out the reification solves all problems and 
provides universal APIs and so on does sound eerily similar to the 
naive enthusiasm of the 1950s when people were sure that atomic 
lettuce was going to save the free world.

Part, if not all, of the problem with 'reification' is that this word 
has, until recent times, been used to refer precisely to this tricky 
and potentially dangerous usage where a formalism is given this odd 
capability of referring to its own expressions. (I'm not saying it 
cannot be done, only that it requires great care to do it without 
completely destroying the semantics of the language. It certainly 
cannot be done without being careful about things like the 
use/mention distinction, names versus references, etc.) However, I 
suspect that Mike is right in that most of the proposed usages of 
what is called 'reification' in RDF are in fact not really 
reification at all: they do not really involve meta-reference, 
quotation or truth-predicates (the dangerous stuff). Much of the 
enthusiasm for RDF, for example, seems to be based on the fact that 
it is easy to use RDF triples as a kind of universal datastructure 
modelling primitive, which is both true and has absolutely nothing to 
do with reification.

>>One way of geting what you want would be to be able to tag real statements
>>somehow.  This is NOT reification, but I don't think that tagging needs
>>reification at all.  Using reification for tagging is like using the entire
>>waste heat output of a malfunctioning nuclear power plant to heat your
>>home.

:-)

>>Peter Patel-Schneider
>
>I think "tagging" in some form is a crucial thing for the language 
>to have - as we move to rules, I think it wil be crucial (I will 
>want to know where rules come from, and maybe use that to break the 
>inevitable A->B, B->C, C->A loops that might arise from distributed 
>rule definitions).  The semantics of SHOE (not as formal as DAML) 
>focused a lot of effort on claims, and we make use of them in a 
>number of ways.

Ah now, THAT is a topic worth discussing. Yes, I agree; we need to be 
able to 'tag' assertions with information about the assertion like 
where it is from, when it was recorded, who vouches for it, whether 
it is trustworthy and who says so and on what grounds, stuff like 
that (not just with extra data fields, as Mike was saying). This is 
'meta' information, so its like reification in a sense, but it's not 
the kind of meta-description that logical reification is concerned 
with, the kind that gives all the selfreferential kind of semantic 
dangers, because it isnt concerned with logical truth. This is going 
to require a semantics all its own, which as far as I know hasnt been 
done yet. (If anyone knows different, please send me pointers!)

> I also agree w/Peter that we may not want to buy into the full RDF 
>reification within the "logical interpretation" aspects of 
>DAML+OIL+?x

One way to do it coherently would be to treat reified RDF triples as 
a concrete data type, a route I think we should deliberately avoid 
taking.

Pat Hayes

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


This archive was generated by hypermail 2.1.4 : 04/02/02 EST