is a missing triple a constraint violation?

What follows is partially a question of terminology, partially
a question of the RDF Schema Working Group intent, and partially
a query about what the community thinks is most useful in regard
to the RDF Schema specification.

On the subject of constraint violation and "inconsistent models",
http://www.w3.org/TR/1999/PR-rdf-schema-19990303/#constraints
[Section 3, "Constraints"] defines the term "inconsistent model"
to describe an RDF model that violates a constraint as specified
by the RDF Schema properties.

The spec explicitly leaves it open to implementations to decide
how to behave in the event of a constraint violation.

But the spec doesn't specify whether a missing triple is a
constraint violation.  After researching the WG mail archives,
I am still not sure whether this omission is intentional or
unintentional.

Early in the WG discussion Andrew Layman and Guha observed that
all the constraints can be interepreted in one of two ways:
http://lists.w3.org/Archives/Member/w3c-rdf-schema-wg/1998JulSep/0083
(sorry, some of the links will be to W3C Member-only mail archives.
I can't retroactively change the access permissions, though I
encourage the authors of the cited messages to repost them to the
IG if they wish.  My summaries that follow should be adequate to
understand the messages, however)

1. as rules that permit inferencing; e.g. to infer a type for the
   subject of a statement from a domain property of the predicate
   of that statement or
2. as integrity constraints; e.g. a graph that does not contain
   a type property where a domain constraint would require one
   would be regarded as being in error in some manner.

This became #26 on the working group's issues list.
http://www.w3.org/RDF/Group/Schema/openissues.html#c26

At the time the working group considered this question there were
many more constraint properties contemplated (minimum and maximum
number of occurences, value constraints, etc.).  Subsequently the
working group decided to limit the constraint properties in the
first version of RDF Schema to the barest minimum.

Issue 26 was decided by adding the definition of "inconsistent
model" that is cited above.  
http://lists.w3.org/Archives/Member/w3c-rdf-schema-wg/1998JulSep/0130
Andrew noted that there may be interoperability issues for
applications that attempt inferencing and suggested that the WG
might need to further discuss this question later.

The working group chose not to directly tackle the question of
inferencing.  Some of us had the expectation that a subsequent
working group would be established to work on an "RDF Logic" layer
and therefore such matters should be deferred to that working group.
Thus, as Sean Luke points out in
http://lists.w3.org/Archives/Public/www-rdf-interest/1999Dec/0117
RDF Schema is (intentionally) weak in specifying inferencing.

As described in
http://www-diglib.stanford.edu/diglib/ginf/rdf-schema-in-ginf.html
(cited by Dan Brickley in
http://lists.w3.org/Archives/Member/w3c-rdf-schema-wg/1999JulSep/0007 )
GINF has implemented validation for RDF Schema, checking that subjects
have types according to domain constraints.  However it also notes
"[it] would be helpful for developers to have a standard set of
inference rules for RDF schema implementation."  Further comments
from the GINF folk would be welcomed.

Recently in a local discussion, the question was raised as to whether
the language of http://www.w3.org/TR/1999/PR-rdf-schema-19990303
precludes an application from performing such inferencing.  I believe
that it does not.  The term "inconsistent model" may be confusing;
it may bring in too many other connotations that are not in the
specification.  In the case of violation of a constraint such as
[2.3.3] "A property can never be declared to be a sub-property of
itself, nor of any of its own sub-properties" this is a reasonable
term, but in the case of a missing triple it may interfere with
peoples' understanding.

Did the Working Group intend to say that a model that does not
contain a triple where one should be expected is also an
"inconsistent model"?  If so, is it appropriate to add words
to the specification to note that in the case of a constraint
violation such as this an implementation might choose to infer
the triple?  I think this would help clarify the intent of
"inconsistent model" and help set the foundation for an
"RDF Logic" layer to be specified.

Comments welcome, especially as Dan Brickley is putting the final
editorial changes into a new RDF Schema Proposed Recommendation.

RDF Schema WG members are encouraged to include
mailto:www-rdf-interest@w3.org in their replies, noting this is
a public forum.

-Ralph

Received on Wednesday, 5 January 2000 11:17:33 UTC