Re: Review of WOWG Semantics editors draft 2003-01-02

> Subject: Review of WOWG Semantics editors draft 2003-01-02
> From: Dave Beckett <dave.beckett@bristol.ac.uk>
> To: "Peter F. " Patel-Schneider <pfps@research.bell-labs.com>,
>         Pat Hayes <phayes@ai.uwf.edu>, Ian Horrocks <horrocks@cs.man.ac.uk>,
>         public-webont-comments@w3.org
> Date: Sat, 04 Jan 2003 15:56:16 +0000
> 
> 
> Review of WOWG Semantics draft
> 
> Comments on
>   http://www-db.research.bell-labs.com/user/pfps/owl/semantics/
> 
>   Dated 2 January 2003
> 
> (CC:ing to authors, not sure if you are all ready for public comments yet)
> 
> I'm not reviewing the detail of the model theory and semantics, not
> my area.
> 
> Summary
> 
> Needs many document fixes and improvements and in particular some of
> the syntax used I feel is confusing.

Summary of changes:

I make a lot of stylistic changes which I hope help matters here.
I've added a bunch of links.  I'm going to add more links.

> General comments
> 
> Please consider making this one HTML document version; the splitting
> over multiple pages makes it difficult to look at all the content
> properly.  Please consider making the single document version the
> only version, or at least the primary document.

I was asked to make this a composite document. There is a
non-normative single-file version.  The single-file version is
machine-generated from the multiple version so they do line up
correctly.  I have also added next (and previous) links to the bottom
of the pages.

> Lots of references to terms in other specifications without links.  I
> point out some of them but this needs a thorough update.  This is to
> aid people using the document, they have to be able to find what you
> are refering to.

This is in progress.  I've added a bunch of links already.  Some of
the links that I want to add are problematic, as the RDF Core WG
documents are not in their final location.

> There are very few links between the syntax.html / semantics.html /
> rdfs.html pages - this tends to indicate to me that they were written
> separately and aren't yet fully integrated as parts of a single
> document.

I believe that the sections are integrated.  There are too few links,
however, so I have added some, and will add more.

> I would ask you also add internal anchors and links each time you use
> terms such as OWL/Lite, "the full abstract syntax", "class",
> "property IDs" etc. and for every definition of a term and use such
> as <axiom>, <classID>.

I've added a lot of these.  I probably need to add more, which will be
done in subsequent passes.

> There are many definitions, of which many have no HTML anchor that
> can be pointed at and would benefit from a consistent markup
> convention so that they are easier to find when reading.

I'll add indications of the important definitions.

> Section Index of Vocabulary
> 
> Please add a statement to indicate what this section is for.  Does it
> define the lists or do the linked sections do so.  In case of error,
> is it definitive?

Paragraph added.  This table has also been tagged as informative.

> Section 2 Abstract Syntax
> http://www-db.research.bell-labs.com/user/pfps/owl/semantics/syntax.html#2
> 
> P2 Is this the editors, WG opinion or normative content?  Does any
> implementation of OWL have to believe that there is a problem with
> RDF graphs?  I think this is inappropriate.

This paragraph was removed.

> "RDF graph" link

Will do.  I need to wait on Concepts being published.

> P3 "a version of Extended BNF"
> Having yet another EBNF is a bad idea, and this one is particularly
> inappropriate.  I suggest you use the one in the XML specification
> which is properly defined http://www.w3.org/TR/REC-xml#sec-notation
> The biggest problem is the use of <> for non-terminals, more of which
> later.

I changed over to the XML way.  It is not as nice to look at, however.
I now think that the benefits later in the paper, most of which you
point out, more than balance the reduced readability of the productions.

> P5 "OWL/Lite"
> This isn't defined till 2.3.1 - please make a link.

Done.

> "RDF Schema" link

Will do (or maybe done already).

> P6 "OWL/DL" Isn't defined till later, please make a link.

Done.
 
> P7 "by means of the canonical URI reference for the datatype,
> http://www.w3.org/2001/XMLSchema#<name>"  needs more explanation.
> <name> is never defined; some minor rewording should be sufficient.

Some changes made.

> The links given are to the definition of the datatypes in XSD,
> not the URI-references of the datatypes.  That seems slightly
> confusing too.

I think that it is better to point to the definition.

> No references to XML Schema, or XML schema data type specs as
> documents.

Added.

> You seem to be using the RECOMMENDED etc words with significance.
> Point to the relevant RFC if that is how you are using them.

Changed.

> Section 2.1
> 
> Here is the first example of the problems with your BNF, you allow
> alternatives to be given by using | or by repeating productions and
> this first set of productions (unlabelled) uses both.  Surely one of
> these is redundant.

There are reasons to not put all the productions for a nonterminal
together.  It is also useful to use |.  I've changed to uniformly
using | where the productions are together. 

> "each of which can have an ID"
> an identifer?  Is ID some term/terminal in the abstract syntax,
> please explain fully.  I will read assuming ID is a synonym for
> identifier, although things like "the foo with ID bar" seems odd
> since all global names are URI-refs.

Changed to identifier.

> "qualified names ... as is done in RDF" - you need to give a pointer
> to this term in XML.  And you might consider pointing at how they are
> used in the RDF/XML syntax, not RDF.  RDF the language has no
> qualified names.

This sentence was removed.

> "to an XML Schema datatype" => add a reference

Do you mean a reference to the discussion in the document itself?

> "If a URI reference is a datatype, i.e., the URI reference points to
> an XML Schema datatype, as described above"
> This sentence seems to imply all datatypes are XML Schema datatypes

Modified.

> If you want to define OWL datatypes, please do so explicitly.

Done.

> "In OWL, as in RDF" => add a reference

Will do.

> "The class with ID owl:Thing" 
> an example of the oddness of ID as I read it.  Surely that is just 
> the same as "The class with identifier/URI reference owl:Thing" or just
> "The class owl:Thing"

I changed "ID" in the text to "identifier".

> Section 2.2
> 
> "striped RDF/XML syntax without the use of rdf:nodeID."
> I'd rather you didn't use the striped word, 

Done.

> but in any case, you do
> need references here to both the syntax and the rdf:nodeID.

Will do.

> Here you use lots of brackets in <propertyValue> rather than just
> repeat the terminal, as you did in 2.1 for <directive> and
> <annotation>.  I really think picking one style only is better.

See above.

> "the full abstract syntax types can be general descriptions, "
> Whoa.  Where did that come from?   

``full'' is used here only to distinguish from the Lite portion.  I've
changed the wording.  

> Is there a section for the "the
> full abs syntax", or is this Section 2?  Does it have a name?  Full OWL,
> OWL?  If so, please define and use it when it was introduced in
> section 1.

This is confusing, so I'm changing to OWL/DL throughout.

> Section 2.3.2.4
> 2nd <axiom> defn, ObjectProperty's ( is on a new line, it would read
> better as ObjectProperty( on one line which is how all the previous
> forms were declared.  The indenting is rather difficult to read too

The indenting was improved.

> Overall Section 2:
> 
> Consider adding a summary table of all the defined terms, pointing to
> the sections of 2.x in which they are defined.

This is in the vocabulary index.

> I'd also suggest tables with complete lists of the terms in the OWL
> variants.

This is in the vocabulary index.

> Section After 2.3.2.4
> Right here I want to read more, but there is no link to the next
> section, in a different document.  Please consider merging these
> documents..

The document is split by section.  I'll add links to the end of each
section.

> Starting again at
> http://www-db.research.bell-labs.com/user/pfps/owl/semantics/direct.html#3

> Section 3
> "RDF graphs", "RDFS model theory"
> Intro needs pointers to the cited documents.

Will do.

> Section 3.1
> Here the BNF choices start to look a bit of a problem
>   "a 3-tuple <Ld,Vd,L2Vd>," 
> - so now <> also means tuples.

The syntax change makes this less a problem.

> "XML Schema" => reference
> " http://www.w3.org/2001/XMLSchema#<name>, " => reference
> plus here what are <name> used as?  tuples? or grammar terms?

Explained.

> "owl:Thing" is not a URI reference, or at least not as given
> here.  Refer to where that terminology is defined.

Fixed, I think.

> Section 3.2
> Give the table a <caption>, link and use it in other sections
> (ditto for 3.3)

Done. Done. Will do as appropriate.

> Make every abs syntax term used here a link to its definition, which
> might be easier if section 2 had a summary table.

Will be done.

> Maybe change the column 1 title to "Abstract Syntax"

Done.

> Some syntax terms in this table are never defined such as unionOf,
> part of <description>.  I looked for it and only found it in 2.3.2.2
> by reference.  The name hints at what it could be, but it is not
> defined in that section except by the words "set of individuals".
> This needs fixing.
> 
> This section uses <> as tuples in the EC(S) column.
> 
> I'd suggest expanding EC(S) as the column header to the full name;
> there is no need for an abbreviation here.

Done.

> Section 3.3
> 
> Column 1 is labelled Directive, it should point to where it is
> defined in 2.1 

Will do.

> "Class(c complete annotation(...) ... "
> 
> would be so much clearer if it pointed to the <axiom>, showed that
> complete was one of two allowed <modality> and that the remaining
> terms were <annotation> followed by <super>

Will do.

> ObjectProperty seems to be different from the <axiom> declarations in
> section 2 which make Functional etc. alternatives; here they are all
> allowed, but at most once each.
>
>   for example 2.3.1.3: <axiom> allows OWL/Lite:
>    ... [Functional | InverseFunctional | Functional InverseFunctional | Transitive] 
> 
>   here:
>     [Symmetric] [Functional] [InverseFunctional] [Transitive]
> 
> which allows all four to be present

The semantics table makes it hard to do the exactly correct thing.
This is close, in that it covers all the syntax, and gives the correct
meaning to all axioms.

> Section 3.4
> 
> the first sentence desperately needs links to the abs syntax, axions etc.

Will do.

> These look like important definitions, worth drawing out of the
> formatting and saying so like:
> 
>   [Definition]: An Abstract OWL Interpretation

These have been more-heavily emphasized.

> Section After 3.4
> 
> again, no link to the next section

Added.

> http://www-db.research.bell-labs.com/user/pfps/owl/semantics/mapping.html#4
> 
> Section 4
> It has some links to the references, but the first paragraph needs
> the rest added.
> 
> P2 the "triples syntax"[sic] also has a document, part of the rdf
> test cases.  Cite it, and it is called "N-Triples".

I have been told not to refer to N-Triples, as they are only
internal.  I'm using the syntax from the RDF MT, which is different.

> The explanation on expansion of the prefixes is inadequate, need to
> explain a little more.

I've added some stuff.

> P4 Here <> are nonterminals again; sec 3 they were tuples, sec 2,
> nonterminals.

The new EBNF helps.

> P5 "0." - typo?

I'm not sure what this was, but I can't find any "0." now, so I guess I
fixed it. 

> P8 new terminology, <> is a document reference URI.

I'm using a different thing for this now.

> Table; give it a caption, link, use it when you reference it in the
> words before it.

Done.

> The column 1 heading is better here; is "Syntax - S" in section 3.

...

> column 1 - <thing...> are non-terminals
> column 2 - <> are document reference URI
>            <thing> are non-terminals
>            or maybe? N-triples URI references such as in:
>               "<> <URI reference> "<URI reference> ."
>            which isn't correct N-Triples syntax either way you use it.
>            See <type1> etc in Individual for example of the
> 	   confusion.  

All <> are now gone, except where they are additional parts of the
transformation.

> Actually first time looking at it I thought there seemed to be
> triples with 4 items in the Individual column 2 but I finally worked
> out that <annotation> was not refering to Annotation(..) higher up,
> but to annotation() further down the table - this is confusing.
> The latter still gets the N-Triples syntax wrong, a URI-reference in
> N-Triples has <>s around the URI-reference.

This should be somewhat better now.

> column 3 - <thing> are syntax terms  
>            or bits of N-Triples such as : 
>               "<literal>"^^<datatypeID> 
>            which is actually wrong, it should be:
>               "<literal>"^^<<datatypeID>>

Fixed.

> This is a big mess I feel and should be addressed by changing some of
> the terminology.  I suggest changing the BNF somewhat and maybe not
> using <> for tuples (although that isn't used in this section), or
> use typographic conventions to help.

Done.

> Section 4.1
> 
> Please give the table some headings, a caption and point to it in the
> text better than "the triples above".

Done.

> Lots of definitions here, which might benefit from typographic
> conventions to highlight them.  

I don't believe that these definitions need more emphasis as they are
only used in this section.

> The triples in-line in sentences
> don't read too well, I'd suggest pulling them out:
> "of the form  x owl:oneOf rdf:nil . or  x owl:oneOf is . where i"

I'll probably use a particular style to show them off better.

> The variable? names 'x', 'y', etc. as used in triples should be made
> clear that they will be replaced with something else in the triple.

Will do.

> Some of the defintions use a list of conditions.  Make it clear
> these all must be satisfies.

Where this is so, I've made sure that there is explicit wording.

> "A quick incomplete gloss of the above is that"
> ??
> 
> Is this paragraph of much use?

This is the wording that I expect to be used elsewhere.  I'm including
this here just as a general indication of what is going on.

> General Section 4
> 
> section 4 is pretty much absent of links after paragraph 2 and this
> is important to fix to relate this with the other sections,
> documents, references and other material

More links will be added.

> http://www-db.research.bell-labs.com/user/pfps/owl/semantics/rdfs.html
> 
> Section 5
> 
> Lots more references needed to XML, RDF, RDFS, the RDF namespace, RDF
> vocabulary, RDF Semantics (not model theory), ...

Will do.

> Section 5.1
> 
> OWL/Full appears here for the first time.   I expect this should be
> defined elsewhere and used throughout these documents.

The semantic definition of OWL/Full is here.  A good place to point
would be to the Reference document.

> Section 5.2
> <> are tuples again

The changes to the EBNF result in many fewer <>s.

> Personally, I find these tables hard to read.  Maybe they should be
> indented so that the sort-of sub-sections "Relationships between OWL
> classes" etc. are clearer, and the tables of conditions are more
> associated with them.

I've put some text around them.

> Sections 5.3.1, 5.3.2
> More important defintions that would benefit from typographical
> emphasis.

I'll consider this.

> http://www-db.research.bell-labs.com/user/pfps/owl/semantics/proofs.html
> Not reviewed in detail, would be good if it linked to the therorems
> that the proofs correspond to.

Will do.

> http://www-db.research.bell-labs.com/user/pfps/owl/semantics/examples.html
> 
> Appendix B
> 
> Are these in the OWL test cases?  Have they been machine checked?

Unknown. Probably not.  They will be made test cases.

> B.2
> Links to OWL/DL, OWL/Full, thanks.

Done.

> References
> 
> I expect some more might be needed here such as the RDF XML Syntax draft  :)

Several have been added.


Thanks again for your review,

Peter F. Patel-Schneider
Bell Labs Research
Lucent Technologies

Received on Tuesday, 14 January 2003 03:07:12 UTC