OWL Implementations as of December 2003 (Historical)


This is based on the Request for CR (by Connolly and Hendler), edited by Sandro Hawke. New information during CR is expected to be incorporated.

Implementations of OWL Technology

There is now an RDF-based dynamic summary of results for systems using the OWL Test Suite.

The following implementation experience leads us to believe that once the exit criteria below are met, we will have sufficient implementation experience to validate the design and merit widespread deployment.

Exit Criteria

When the Working Group requested CR, it suggested the following conditions met before proceeding on to PR. In terms of W3C Process, this is the immediate implementation goal, but of course the real goal is to have available lots of excellent, interoperable, and downright useful OWL systems. These criteria are reproduced here so they can be linked to appropriate implementations as they emerge.

  1. finish resolving dependency on RDF Core specs, esp. RDF Semantics

    note 10 Oct RDF last call drafts

  2. two complete OWL Lite consistency checkers (i.e. 2 which pass almost all OWL Lite consistency and inconsistency tests and moreover claim logical completeness)

    candidates include Pellet and Hoolet.

  3. Each test (except the extra credit tests) is demonstrated to be passed by some implementation

    stay tuned to OWL test case results

  4. two reasoners implementing (different) substantial subsets of OWL DL

    candidates include FaCT, Racer, Cerebra

  5. two reasoners implementing useful subsets of OWL Full and passing at least 80% of the postive entailment tests

    Candidates include Euler, Jena, and surnia

  6. two owl syntax checkers passing all tests

    candidates include Cerebra, OWLP, OWL validator above

Features at Risk

One technical detail concerning structure reuse in OWL Semantics and Abstract Syntax, Section 4 (Mapping to RDF Graphs) has been identified as "at risk" and subject to change. We expect this change, if made, to simplify rather than complicate implementations, and since it is a relaxation of a current restriction, it will not invalidate or change the meaning of any valid OWL or RDF documents.

The simplification is sketched in sections B.1 and B.2 of a proposal from a 4 March editor's meeting and elaborated somewhat shortly after...

We add to the preamble to the mapping rules words like:
[bnode reuse]
"When processing an abstract syntax construct corresponding
to either the description, restriction or dataRange construct
then, if that exact instance of the construct has already
occurred then there is at least one blank
node already corresponding to the construct. In such a case,
the mapping may nondeterministically use any previous result
(a blank node) or may apply the mapping rues to the new occurrence."
JJC 4Jun

The following tests would change from being OWL Full (in)consistency tests of OWL Full files, to being OWL Lite and OWL Full (in)consistency tests of OWL Lite files:

The following tests would change from being OWL Full (in)consistency tests of OWL Full files, to being OWL DL and OWL Full (in)consistency tests of OWL DL files:

The following similar tests would be unchanged (in OWL Lite or OWL DL):

The following similar tests would be unchanged (in OWL Full):

Change Log

Constructed this document out of the Request for CR. (sandro@w3.org)
Added at-risk section, mostly from the Request for CR. (connolly@w3.org and sandro@w3.org)
Added link to test results page. (sandro@w3.org)
integrated some implementation feedback: claims of completeness, etc. (connolly)

Sandro Hawke
$Id: impls.html,v 1.18 2004/12/06 13:27:54 sandro Exp $