Peter noted that RDBMSs are restrictive with respect to properties, while DAML+OIL is permissive.
The Stanford JTP implmementation supports checkpointing with rollback, and uses this to handle premises.
In general, premises are useful for accessing subsumption style reasoning with variable bindings. They can also be used for "what if" analyses.
subclassOfrelations is critical for software tools and query relaxation. The RQL direct operator (^) is equivalent to a query with negation: "is X a subclass of Y and is there no other class that is an intermediate subclass between X and Y".
The direct operator can introduce non-monotonicity in an open world, as noted earlier. Pat suggested that having non-monotonic constructions in the query language is probably OK, but non-monotonic conclusions are not.
Richard noted that identifying direct relationships was needed in OKBC. He suggestd that we might want to separate structural and entailment queries.
There was no consensus on supporting the direct queries.
ACTION (Ian): investigate complexity of disjunction and atomic negation
select distinct, eliminating duplicate bindings to reduce client and network overhead. In some cases, you want duplicates; in others, you don't. For example, given the query "what books did Pat buy?", you might or might not be concerned that the same book was purchased twice.
Peter noted that this is like DBMS projection, and gets even more complicated with inference.
Richard noted that DQL currently returns unique sets of bindings for all variables mentioned in the query.
RESOLVED: DQL should provide some mechanism for returning subsets of the variables mentioned in the query. The user can get multiple occurances of the same bindings for some variables by adding the variables occuring along multiple distinct paths.
ACTION (Richard): summarize discussion in an email message.
subClassOfcycles. They haven't explicitly addressed
subPropertyOfyet, but are expected to follow suit.
There has also been some discussion about whether
is to be interpreted as a string or as a QName.
"Quick fix" proposals of making the telecon 2 or 4 hours earlier didn't work even for those present.
ACTION (Mike): poll the group for a possible new time
We'll continue the discussion regarding
Several folks (Ian, Dieter, Frank, and Stefan -- Pat and Deb were also invited) will be partcipating in the NSF/EU Semantic Web strategy meeting in France during next week's telecon. We decided to have a telecon anyway. They'll provide a report on October 9.
$Id: 2001-09-25.html,v 1.6 2001/10/02 21:05:00 mdean Exp $