DAML Rules!!!

From: pat hayes (phayes@ai.uwf.edu)
Date: 06/05/01


There are two somewhat different ideas of what a rules langauge 
should be like/for, and Pat and Stefan have been thrashing them out 
and arguing about the distinctions between them. This attempts to be 
a summary of what we talked about.

One idea (call this RULES4DAML) is that the rules should be thought 
of as Horn clauses which preserve validity wrt DAML semantics or 
maybe some extension of current DAML semantics. This seems likely to 
lead to languages similar to CARIN (Rousset and Levy, 1996). (which 
Pat was ignorant of when he wrote the earlier strawman proposal, and 
which he agrees needs careful study.)

Another idea (call this GENRULES) is a more general-purpose tool 
which is designed to make it easier to write operations on RDF 
Kbases. This would also use Horn-clause form, but for pragmatic 
rather than DAML-semantic reasons (ease of rule specification, good 
assertional programming practice, etc..). This would be immediately 
useful and be applicable to any piece of RDF, and in particular could 
be used to implement RULES4DAML, but also (for example) the kind of 
transformations into circle-and-arrows graphical forms that Dan 
Connolly mentioned a few days ago.

GENRULES would be simpler to specify than RULES4DAML.  In an ideal 
world, it might be possible to specify RULES4DAML as a sublanguage of 
GENRULES (they both would be Horn clause languages), but in the 
immediate short term they should probably be developed seperately.

Source Identifiers and Skolem Functions

Since RDF data is distributed and different RDF source have different 
it is necessary to be able distinguish between RDF data coming from different
sources. Source or Context identifiers as a building block in the 
DAML rule language
allow to based inferencing based on different sources of RDF data.
A slight generalization of Context identifiers to skolem functions allows to
construct parameterized RDF models.

Negations and monotonicity

RULES4DAML should probably have a strictly monotonic semantics. (?) 
On the other hand, an effective Horn logic programming language (like 
GENRULES wants to be) may well want to work with a closed-world 
assumption and negation by failure, for efficiency reasons. It could 
also then utilize the the Lloyd-Topor transformation which allows to 
have expressive (FOL) rule bodies which can be compiled to ordinary 
horn rules with a CWA based negation.

All of which suggests we need a way to temporarily operate with a CWA 
and report the results to a larger context in a monotonic framework. 
Source identifiers as context labels may provide a way to do this 
cleanly. (This connects with some recent discussions on RDFLogic from 
Lynn, Sandro and others, BTW.)

IHMC					(850)434 8903   home
40 South Alcaniz St.			(850)202 4416   office
Pensacola,  FL 32501			(850)202 4440   fax

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