Timeline - editorial issues with docs - implementation

I am still highly suspicious of the proposed timeline.

I have significant concerns about:

Document related:
- the editorial quality of our docs
  (sufficient to vote against last call)
- lack of anchors in our docs.
- the lack of cross links between our docs
- the lack of links from our docs to the RDF docs
- the lack of clarity as to when specs are informative or normative

Implementation related:
- the lack of implementation of: OWL Lite or OWL DL subset identification.
- test case creation and approval

In more detail:

+ editorial quality
Most of our WDs have a number of informal asides that are not, in my
opinion, appropriate, in the final recs. Also the
e.g.
[[
This syntax does not have to worry about any of the problems induced by the
RDF triple model, including non-closed and ill-formed lists and
restrictions. No parsetype extensions are needed for readability, and many
issues of coordination with the RDF Core WG are not relevant at this level
of syntax. Layering issues can also be safely ignored. Further, namespace
issues can also be somewhat ignored; in the syntax here reserved words are
not given with any namespace qualification
]]
says very little pertinent to a langauge specification.
This reflects a lack of thorough review at that sort of level in the WG,
because, quite rightly, we have been concentrating on the technical meat
rather than presentation.


+ lack of anchors in our docs.
+ the lack of cross links between our docs
In a multidocument recommendation it is important that the reader can
navigate across the documents.
This means that we need to have links from say the direct semantics of
intersectionof, its abstract syntax, its mapping to triples, its feature
synopsis, its treatment in the guide, its triple based semantics. Since
*none* of these have anchors, we probably need to do this in two phases. (i)
put the anchors in (ii) at the cross-links.

+ lack of links between our docs and the RDF specs.
There are a number of goals which we realize through using RDF. e.g. those
to do with internationalization. If we do not link our concept of URIref and
Literal to the RDF recs then we do not achieve those goals, and we have
defined a langauge that is appropriate for deployment in the united states
(only).

+ the lack of clarity as to when specs are informative or normative

e.g.
[[ Properties may be stated to have ranges, (i.e., if X is related to Y by a
property p with a range class, then Y must be an instance of the range
class). For example, the property hasChild may be stated to have the range
of Mammal. From this a reasoner may deduce that if Louise is related to
Deborah by the hasChild property, i.e., Deborah is the child of Louise, then
Deborah is a Mammal. Range is also a global restriction as is domain above.
See the discussion below on local restrictions for more information. ]]
I take this discussion from the features document to be informative, but
that is my take - it is not clear in our recommendation.

e.g.
[[
The semantics here starts with the notion of a vocabulary, which can be
thought of as the URI references that are of interest in a knowledge base.
It is, however, not necessary that a vocabulary consist only of the URI
references in a knowledge base.
]]
Is that intended to mean anything for implementations? Is support for
vocabulary other than that of URI references an OPTIONAL feature for OWL? I
hope not.


I suggest the planned January publication should not be last call but should
be the editors' best efforts at last call ready documents with:
- spurious personal opinion deleted
- anchors added liberally
The main purpose is for an in-depth review by the WG and our friends, and
then we present more highly polished documents as a last call to the rest of
the world, a month or two later.


+ the lack of implementation of: OWL Lite or OWL DL subset identification.

As far as I understand the implementation experience of the group is either
RDF based or DL based. Our documents have the abstract syntax to triples
mapping as a way of bridging these. Since RDF/XML is normative, it is the
reverse transform that is crucial.
All conformant implementation will need to be able to say whether an RDF/XML
document is or is not an OWL Lite document. OWL DL and OWL Full
implementations moreover need to be able to indicate whether an RDF/XML
document is or is not an OWL DL document.
I suspect that once we have such implementations, we may find that the
abstract syntax is substantially more restrictive than we actually want.
(e.g. the use of bnodes is very limited).


+ test case creation and approval

We do not have enough, and we have not spent enough time on agreeing or not
the ones that we do have. (This is partly my fault).


Finally: a process question

We are about to close the last remaining issues - do I need to officially
create some new ones e.g. "ISSUE editorial" to justify voting against a last
call, which I certainly intend to do?

Jeremy

Received on Friday, 6 December 2002 05:22:34 UTC