Joint Committee Minutes 4 September 2001

This page summarizes the telecon for the Joint US/EU Committee on Agent Markup Languages held from 1300 to 1400 PDT on 4 September 2001. These minutes were prepared by the chairman, and were approved during the 18 September telecon.



No announcements this week.

Last Week's Minutes

The minutes from August 28 were approved by those present.

W3C Submission Status

There have been no problems with Lucent's internal review process. Peter will check Thursday for final process completion, then submit to W3C including Deb's cover page updates (alphabetical editors and authors, OIL reference). We'll plan to submit the package as individual documents -- if accepted, W3C will make it a Note. We may have a month or so to make minor editorial updates. The W3C Note submission process is not so well defined; W3C may or may not provide feedback to the submitters.

Dan Connolly re-opened this topic at the end of the meeting. ACTION (Dan Connolly): figure out how to best transfer rights to W3C. If necessary, draft a letter for individual editors and authors to sign.

RDF Core Feedback

Pat Hayes reported on the cyclic subclass discussion during Friday's RDF Core telecon and related email discussions. Most of the RDF Core WG is willing to allow subClass loops to conform to the DAML+OIL suggestion, but a few are still "defending the Java implementors". There's been no final decision yet, pending the return of a member previously strongly opposed to cycles. Pat expects that explicit statement of cycles will be discouraged/deprecated in schema design, but not illegal.

RDFS subclasses need no longer be proper.

Query Proposal

Richard introduced the KSL strawman query proposal.

Richard has also been looking at RQL, which provides a superset of the strawman proposal, including comparisons, non-monotonic operators (e.g. proper/direct subclasses), disjunctions, and negations.

RQL uses ^ syntactically to distinguish between explicit assertions vs. inferences. Forward-chaining reasoners may not keep track of explicit assertions vs. inferences. Requiring this capability may exclude some current reasoners.

Mike noted that conjunction (e.g. returning Californians or Texans) may be a difference between rules and queries. Providing multiple rules with partially common heads may be fine, but people generally want queries to be atomic for performance and other reasons.

A DAML Query should return the entailment based on DAML+OIL semantics.

There was discussion about query controls to support streaming/batching, etc. and allowing the query engine to report that's all, there might be more, or I have have more. Several people felt this was a semantic issue and not just an API/implementation issue.

Ignore the comment about the unique names assumption in the example distributed with the strawman proposal.

Frank noted that RQL is a strategic opportunity to (keep) the database community on-board. Squish, etc. are subsets of RQL, but without any inference/entailment (we might want to also support such queries on only asserted statements as a debugging feature if nothing else).

We could have multiple surface syntaxes on top of a DAML+OIL internal core representation (interlingua).

RQL may already be a W3C submission. ACTION (Frank): verify this, distribute URI.

ACTION (Richard): map strawman proposal to RQL

Richard noted that we need to identify criteria/requirements for a DAML+OIL query language. Ian raised the issue of cycles in variables, discussed more fully in his paper (cited by Richard) and its references.

DAML+OIL Example Updates?

Mike has received several messages noting perceived problems with our daml+oil-ex example (largely the result of running the DAML Validator and other tools): undefined properties and classes, disconnected properties, missing examples (see the list in the agenda for details). There's an opportunity to address these as editorial updates before the proposed W3C Note is finalized (though no one felt we should delay the Note).

Some of the issues can be addressed by documentation in the walkthru. For example, we want to show that it's allowable to use a property without defining it as such (but tools could properly flag this as a possible typo).

We should continue to include at least one instance (e.g. Adam) in the example to show that this is allowed, but may want to move other instances (Ian, Peter, etc.) to another page to show how ontologies and instances are normally separated.


Kelly Barber's raw notes
last week's minutes
Joint Committee home page

$Id: 2001-09-04.html,v 1.5 2001/09/24 21:47:25 mdean Exp $