% notes from JC telecon 12/2/03
% by Benjamin Grosof

agenda: 
- (a bit) feedback on SWRL 
- (mainly) discussion of next steps and directions for SWRL; 

participants:
Mike Dean
Benjamin Grosof
Harold Boley
Ian Horrocks
Peter Patel-Schneider
Deborah McGuinness
Sandro Hawke (joined late)
(regrets from Said Tabet)


%%%%

more feedback on SWRL

misc. questions raised in discussions:

- what are semantics of Datalog

- where to find the canonical definition of RuleML 

%%%%


Mike and Benjamin briefly walked through possible directions
from their emails

Benj suggestions on next steps:

overall, let's start with stuff that's relatively easy and uncontroversial
as well as useful

1. sensor built-ins starting with datatype comparisons and conversions

2. n-ary predicates, which includes predicates not only 
   being OWL classes and predicates
- this is something the OWL and RDF efforts back-burnered
- can do pretty straightforwardly via slots 

wrt n-ary:  
- Mike:  issue of supporting legacy n-ary stuff done in RDF
- Benj:  how would they be any worse off than now, can always treat it as 2-ary

Harold and Deborah and all:  let's be driven by use cases, incl. services

possible work items for future:
- look at use cases of doing n-ary, and of rules for services generally

3. more Lloyd-Topor expressiveness, e.g., 
- OR in the body -- is quite useful for practical purposes
- this is pretty straightforward as syntactic sugar -- it's reducible to Horn
- could also have existentials in the body -- particularly useful with n-ary
 . maybe wait on this til we have more use cases
- could also have universals in the head
 . maybe wait on this til we have more use cases

Peter:  worried that there be monsters here in this territory,
since we're on top of / in combination with full DL not just within Horn,
e.g., may break some implementations' behavior that wouldn't otherwise be
broken; 
e.g., or if attempt later to give an autoepistemic logic kind of semantics
(Benj:  often people talk about this in terms of intuitionistic rather
than autoepistemic);
e.g., if implement in the following fashion:  
query for the antecedent, then if that's satisfied, 
add the consequent with appropriate bindings 
-- since the DL sublanguage may contain [non-intuitionistic]
disjunctive info 

Benj:  if we're using a FOL semantics then it's not problematic;
your example reasoner is just of a
(potentially) incomplete reasoner being (actually) incomplete

%%%%

for next week:

Mike will present more about his suggestions on 
datatype comparisons/conversions builtins

relevant paper to read for next week: 
paper by Ian Horrocks et al on datatype groups
from ISWC-2003, is on Ian's webpage