[pedantic-web] The OWL Ontology URI

[Niklas Lindström]

Hi all,

how come the OWL Ontology itself is defined as
<http://www.w3.org/2002/07/owl>, and not
<http://www.w3.org/2002/07/owl#>? The latter (hash URI, ending in "#")
is linked to, via rdfs:isDefinedBy, for all classes and properties,
but the former is the resource described as the owl:Ontology.

Is this really intentional? It seems to conflate the document and the
ontology. And ontologies aren't considered to be information
resources, are they? Neither classes and properties are, nor is the
thing linked to via isDefinedBy (itself not described further in the
document).

Also notice that neither RDF nor RDFS are described like this -- they
both use the hash URI as identifier for the Ontology (also linked to
with rdfs:isDefinedBy from their classes/properties).

Best regards,
Niklas

[Lin Clark]

Funny, I noticed this over the weekend as well. I was testing a few different ontologies, OWL was the only one I saw that was defined like this.

-Lin

2010/5/5 Niklas Lindström <linds...@gmail.com>

[Richard Cyganiak]

Niklas, please go to http://pedantic-web.org/#rules and look at item
#5. I removed semantic-web from the cc list.

On 5 May 2010, at 19:54, Niklas Lindström wrote:
> how come the OWL Ontology itself is defined as
> <http://www.w3.org/2002/07/owl>, and not
> <http://www.w3.org/2002/07/owl#>? The latter (hash URI, ending in "#")
> is linked to, via rdfs:isDefinedBy, for all classes and properties,
> but the former is the resource described as the owl:Ontology.
>
> Is this really intentional? It seems to conflate the document and the
> ontology. And ontologies aren't considered to be information
> resources, are they?

I think that ontologies are information resources. An ontology is a
document, written down in OWL or another ontology language.

> Neither classes and properties are,

That may or may not be true, but it's besides the point. An ontology
defines (and thereby describes) classes and properties. Information
resources can describe any kind of thing, including classes and
properties.

> nor is the
> thing linked to via isDefinedBy (itself not described further in the
> document).

I agree that this is a problem; the target of the isDefinedBy links
should either be further described in the document, or it should be a
document itself. (rdfs:isDefinedBy is a subproperty of rdfs:seeAlso,
so it certainly makes sense that the target would be a document.)

> Also notice that neither RDF nor RDFS are described like this -- they
> both use the hash URI as identifier for the Ontology (also linked to
> with rdfs:isDefinedBy from their classes/properties).

This is an interesting question, so I did a little study using the
collection of all namespace URIs from http://prefix.cc/ . The goal was
to find out how owl:Ontology and rdfs:isDefinedBy are used in the wild.

346 total namespace URIs
238 are hash URIs
192 could be dereferenced and parsed as RDF (using any23)
105 contain rdfs:isDefinedBy or owl:Ontology

Of the 102 unique resources typed as owl:Ontology, 80.4% are hash-less
URIs, the other 19.6% end in a hash.

Of the 97 unique resources that are targets of rdfs:isDefinedBy
statements,
51.5% are hash URIs *with* fragment (most of them pointing to anchors
within HTML documents),
26.8% are hash-less URIs,
21.6% are URIs ending in a hash.

Finally I looked at those vocabularies where the namespace URI,
owl:Ontology resource, and rdfs:isDefinedBy are either identical or
differ only in the presence/absence of the trailing hash. There were
40 of those documents. Results:

47.5% - Ontology and isDefinedBy target do not end in hash
27.5% - Ontology does not end in hash, isDefinedBy ends in hash
22.5% - Neither ontology nor isDefinedBy target end in hash
2.5% - Ontology ends in hash, isDefinedBy target does not

The detailed lists for each of these groups are attached at the end of
this message.

I conclude that pedants should not use trailing hashes -- neither for
the owl:Ontology typed resource, nor for the target of
rdfs:isDefinedBy triples. This is the popular thing anyway, it means
that only 19.6% of owl:Ontology URIs and 21.6% of rdfs:isDefinedBy
targets needs fixing, and it embraces the consistent view that classes
and properties are defined by ontologies, which are documents.

A number of high-profile vocabularies do not use this approach: OWL,
RDF, RDFS, DOAP, SIOC. But all of these predate the W3C TAG's
httpRange-14 decision, so they were designed at a time when the
interactions between RDF and URIs and HTTP were not yet settled. So
I'd say let's not copy their archaic style, but let's get them to fix
it.

Any disagreement? Bring it on ;-)

Richard



Ontology and isDefinedBy have no hash
http://purl.org/net/ns/wordmap
http://purl.org/net/vocab/2004/03/label
http://purl.org/NET/acc
http://purl.org/NET/lx
http://purl.org/NET/puc
http://purl.org/vocab/changeset/schema
http://purl.org/vocab/frbr/core
http://vocab.deri.ie/am
http://vocab.deri.ie/c4n
http://vocab.deri.ie/dady
http://vocab.deri.ie/dcat
http://vocab.deri.ie/rooms
http://vocab.sindice.com/xfn
http://www.aktors.org/ontology/portal
http://www.aktors.org/ontology/support
http://www.kanzaki.com/ns/music
http://www.openrdf.org/rdf/2009/metadata
http://www.openrdf.org/rdf/2009/object
http://www.w3.org/2004/09/fresnel


Ontology has no hash, isDefinedBy has hash
http://purl.org/net/biblio
http://schemas.talis.com/2005/address/schema
http://schemas.talis.com/2005/dir/schema
http://schemas.talis.com/2005/library/schema
http://schemas.talis.com/2005/service/schema
http://schemas.talis.com/2005/user/schema
http://vocab.org/aiiso/schema
http://vocab.org/lifecycle/schema
http://vocab.org/resourcelist/schema
http://www.junkwork.net/xml/DocumentList
http://www.w3.org/2002/07/owl


Ontology and isDefinedBy have hash
http://demiblog.org/vocab/oauth
http://ontologi.es/lang/core
http://ontologi.es/status
http://purl.org/stuff/rev
http://rdfs.org/sioc/ns
http://rdfs.org/sioc/types
http://usefulinc.com/ns/doap
http://www.w3.org/1999/02/22-rdf-syntax-ns
http://www.w3.org/2000/01/rdf-schema


Ontology has hash, isDefinedBy has no hash
http://purl.org/net/provenance/ns#





>
> Best regards,
> Niklas

[iand]

On May 6, 10:51 pm, Richard Cyganiak <rich...@cyganiak.de> wrote:
> I think that ontologies are information resources. An ontology is a  
> document, written down in OWL or another ontology language.

The view that an ontology is a document is inconsistent with the OWL
Spec which says:

"An OWL ontology is an RDF graph, which is in turn a set of RDF
triples. As with any RDF graph, an OWL ontology graph can be written
in many different syntactic forms..." -- http://www.w3.org/TR/owl-ref/#Syntax

So I believe there is a distinction to be made between the ontology
and the document describing it.

> I conclude that pedants should not use trailing hashes -- neither for  
> the owl:Ontology typed resource, nor for the target of  
> rdfs:isDefinedBy triples. This is the popular thing anyway, it means  
> that only 19.6% of owl:Ontology URIs and 21.6% of rdfs:isDefinedBy  
> targets needs fixing, and it embraces the consistent view that classes  
> and properties are defined by ontologies, which are documents.
>
> A number of high-profile vocabularies do not use this approach: OWL,  
> RDF, RDFS, DOAP, SIOC. But all of these predate the W3C TAG's  
> httpRange-14 decision, so they were designed at a time when the  
> interactions between RDF and URIs and HTTP were not yet settled. So  
> I'd say let's not copy their archaic style, but let's get them to fix  
> it.
>
> Any disagreement? Bring it on ;-)

I think the ontology URI should be the same as the rdf:isDefinedBy
URI. Most of the schemas in your second group were authored by me, or
by people advised by me, but I now believe they are wrong.

I don't have a view about whether that URI should be terminated with a
hash or not. I think that is a completely orthogonal concern.

>
> Richard
>

Ian

[Richard Cyganiak]

httpRange-14 debate ahead! Run for the hills!

On 7 May 2010, at 15:30, iand wrote:
> On May 6, 10:51 pm, Richard Cyganiak <rich...@cyganiak.de> wrote:
>> I think that ontologies are information resources. An ontology is a
>> document, written down in OWL or another ontology language.
>

> The view that an ontology is a document is inconsistent with the OWL
> Spec which says:
>
> "An OWL ontology is an RDF graph, which is in turn a set of RDF
> triples. As with any RDF graph, an OWL ontology graph can be written
> in many different syntactic forms..." -- http://www.w3.org/TR/owl-ref/#Syntax

This doesn't disagree with the view that an ontology can be published
as a document. Quite the opposite.

The passage confirms that an "ontology" is a kind of "RDF graph".

An "RDF graph" can be serialized without loss in, let's say, N-
Triples. In fact, you can trivially round-trip between them.

On the web, "document" ~= "information resource".

"information resource" is defined as: "a resource whose essential
characteristics can be conveyed in a message." [1]

Since the N-Triples serialized form conveys every aspect of an "RDF
graph" or "ontology", they meet the definition.

To put it another way: An N-Triples serialization of an RDF graph is a
perfect representation of that graph. The fact that you can round-trip
between them makes this clear. If it can have a representation, then
it's an information resource and therefore it can be published as a
web document (with 200 status code that returns the representation).

> So I believe there is a distinction to be made between the ontology
> and the document describing it.

The document doesn't *describe* the ontology, like you would describe
a person or tree or a fuzzy concept. The document *serializes* the
ontology, triple for triple, in the same way that a png file
serializes a pixel image. The graph does not describe the ontology;
the graph *is* the ontology, per your quote above.

<snip>

> I think the ontology URI should be the same as the rdf:isDefinedBy
> URI.

+1

> Most of the schemas in your second group were authored by me, or
> by people advised by me, but I now believe they are wrong.

Good to hear that. Any chance of getting these schemas changed, in the
mid-term?

> I don't have a view about whether that URI should be terminated with a
> hash or not. I think that is a completely orthogonal concern.

Yes, although the question wether an ontology is the kind of thing
that can be published as a web document has some bearing.

Cheers,
Richard

[1] http://www.w3.org/TR/webarch/#id-resources



>
>>
>> Richard
>>
>
> Ian

[Antoine Zimmermann]

Richard,


I think that what Ian said is in fact correct and that your argument
does not defeat Ian's.
You can say "serialise" instead of "describe" if you want, but it does
not change the problem. A serialisation of a graph *is not* the graph.

For the sake of an example, let's stick to RDF only. Consider the following:

I have an RDF graph G = "http://ex.org/graph#" (it is a set of RDF triples)
An RDF/XML document at http://ex.org/graph.rdf serialises G.
A Turtle document at http://ex.org/graph.ttl serialises G.
Another RDF/XML document at http://ex.org/graph-v1.0.rdf serialises
graph G (e.g., the names of the blank nodes changed, which does not
change the graph at all).
etc...

They are all information resources but are not *the* graph.

Same thing for ontologies. Moreover, there other arguments in favour
of Ian's view, in the OWL spec, more particularly in OWL 2.
In OWL 2, ontologies can have a version IRI which is used to identify
a specific version of the ontology. Different versions would be
described in different documents or files, therefore, different
versions should be described in information resources having different
URIs.
However, the ontology itself must have the same URI throughout its
evolution, version after version. So you have information resources
that do not necessarily match the ontology URI.

Additionally, though less important, notice that in OWL DL,
ontologies, classes, annotation properties, object properties,
datatype properties and instances are pairwise disjoint. So, an
ontology that is an instance of, say, foaf:Document is automatically
out of the OWL DL profile. It would be too bad that DL-compliant
ontologies would be automatically non-pedantic-proof!

For all these reasons, I would encourage treating ontologies as
non-information resources, well distinct from the file that contains
their serialisation.
E.g., you can start you RDF/XML ontology like this:

<rdf:RDF>
...
<owl:Ontology rdf:about="http://e.org/myOnto#">
...
<owl:versionIRI rdf:resource="http://e.org/myOnto-beta1.owl"/>
...
</owl:Ontology>
<foaf:Document rdf:about="">
...blabla
</foaf:Document>

etc..

----

As a side note, it should be noted that the RDF, RDFS and OWL
vocabularies are described and fully specified in their respective
recommendations. So making use of the so-called RDF/RDFS/OWL ontology
only makes sense if one *does not* implement the said specifications.
Yet, using these files, which are written in RDFS, without
implementing RDFS is quite pointless.

I would like to know if anyone has an example of use cases for these
"ontologies"? I can't find any, or maybe I can foresee *extremely*
limited utility of these files.

> Any chance of getting these schemas changed, in the mid-term?

I can imagine that the RDF and/or RDFS ontologies will be updated when
the RDF working group will be finalising the next version of RDF. I
cannot guarantee anything for the OWL ontology.


Regards,
--
--AZ

[Richard Cyganiak]

Antoine,

On 7 May 2010, at 19:03, Antoine Zimmermann wrote:
> A serialisation of a graph *is not* the graph.

That's correct, but has nothing to do with what I said.

I said that a serialisation of a graph makes an excellent
*representation* of the graph, in the WebArch/REST sense. I believe
this to be trivially true.

More to the point:

"An information resource is a resource whose essential characteristics

can be conveyed in a message."

What characteristic of an RDF graph is it that cannot be conveyed in a
message?

> For the sake of an example, let's stick to RDF only. Consider the
> following:
>
> I have an RDF graph G = "http://ex.org/graph#" (it is a set of RDF
> triples)
> An RDF/XML document at http://ex.org/graph.rdf serialises G.
> A Turtle document at http://ex.org/graph.ttl serialises G.
> Another RDF/XML document at http://ex.org/graph-v1.0.rdf serialises
> graph G (e.g., the names of the blank nodes changed, which does not
> change the graph at all).
> etc...
>

> They are all information resources but are not *the* graph.

Well, if you're the owner of ex.org, then you're free to tell me what
your URIs identify, and there is no violation of web architecture in
what you said.

However, if I'm the owner of ex.org, then I can also have my own graph
G, which is again a set of RDF triples, and I can assign the URI "http://ex.org/graph
" to it, and configure my server to respond to HTTP GET on that URI
with an RDF/XML, Turtle, or N-Triples serialization of the graph,
depending on content negotiation, using a 200 OK status code.

In doing so, I have committed to the following assertions:

1. "http://ex.org/graph" is a graph, a set of triples.

2. "http://ex.org/graph" has three representations, one each in RDF/
XML, Turtle, and N-Triples.

3. "http://ex.org/graph" is an information resource (because of the
200 OK status code), and hence a document.

Where is the supposed contradiction here?

> In OWL 2, ontologies can have a version IRI which is used to identify
> a specific version of the ontology. Different versions would be
> described in different documents or files, therefore, different
> versions should be described in information resources having
> different URIs.
> However, the ontology itself must have the same URI throughout its
> evolution, version after version.

s/described/serialized/ please. A document containing an RDF graph
doesn't describe that graph, it serializes it. (Or: encodes it.)

Other than that, what you say above is fine, let the ontology have the
same constant URI throughout its lifetime. But how does this prevent
the ontology from being an information resource?

> Additionally, though less important, notice that in OWL DL,
> ontologies, classes, annotation properties, object properties,
> datatype properties and instances are pairwise disjoint. So, an
> ontology that is an instance of, say, foaf:Document is automatically
> out of the OWL DL profile. It would be too bad that DL-compliant
> ontologies would be automatically non-pedantic-proof!

There is any number of things that are trivially true but are illegal
to say in OWL DL. If you care about it, then just don't assert that a
resource is both an owl:Ontology and a foaf:Document. It does not
follow that ontologies are not information resources.

> For all these reasons, I would encourage treating ontologies as
> non-information resources, well distinct from the file that contains
> their serialisation.

There are no "files" in web architecture. I encourage you to read up
on the distinction between resource and representation:

http://en.wikipedia.org/wiki/Representational_State_Transfer#Guiding_principles_of_the_interface
http://www.w3.org/TR/webarch/#internet-media-type
http://www.ics.uci.edu/~fielding/pubs/dissertation/rest_arch_style.htm#sec_5_2
http://lists.w3.org/Archives/Public/www-tag/2005Jun/0039

Ontologies are certainly distinct from their representations.

> As a side note, it should be noted that the RDF, RDFS and OWL
> vocabularies are described and fully specified in their respective
> recommendations. So making use of the so-called RDF/RDFS/OWL ontology
> only makes sense if one *does not* implement the said specifications.
>
> Yet, using these files, which are written in RDFS, without
> implementing RDFS is quite pointless.

To use these files, you need to implement RDF/XML and the RDF data
model. You don't need to implement RDF's built-in terms, or RDFS, or
OWL. There are many tools that meet this description, for example many
SPARQL processors.

> I would like to know if anyone has an example of use cases for these
> "ontologies"? I can't find any, or maybe I can foresee *extremely*
> limited utility of these files.

Looking up a list of the terms that are defined in the rdf/rdfs/owl
namespaces. Looking up their rdfs:labels and rdfs:comments.

The core classes and properties of the web of data should resolve to
RDF descriptions of themselves just like any other class and property.

Best,
Richard

[Stuart A. Yeates]

On Sat, May 8, 2010 at 9:57 AM, Richard Cyganiak <ric...@cyganiak.de> wrote:
> 2. "http://ex.org/graph" has three representations, one each in RDF/XML,
> Turtle, and N-Triples.

2. "http://ex.org/graph" has AT LEAST three representations, ...

I'm no expert on turtle or n-triples, but I can tell you that there
are an infinite number of ways to encode an RDF graph in XML, and
further, while the semantic web world chooses not to differentiate
between those encodings, there are other uses of XML for which those
differences are differentiated and for which those differences matter.

cheers
stuart

[Kingsley Idehen]

Richard Cyganiak wrote:
> httpRange-14 debate ahead! Run for the hills!
>
> On 7 May 2010, at 15:30, iand wrote:
>> On May 6, 10:51 pm, Richard Cyganiak <rich...@cyganiak.de> wrote:
>>> I think that ontologies are information resources. An ontology is a
>>> document, written down in OWL or another ontology language.
>>
>
>> The view that an ontology is a document is inconsistent with the OWL
>> Spec which says:
>>
>> "An OWL ontology is an RDF graph, which is in turn a set of RDF
>> triples. As with any RDF graph, an OWL ontology graph can be written
>> in many different syntactic forms..." --
>> http://www.w3.org/TR/owl-ref/#Syntax
>
> This doesn't disagree with the view that an ontology can be published
> as a document. Quite the opposite.
>
> The passage confirms that an "ontology" is a kind of "RDF graph".
>
> An "RDF graph" can be serialized without loss in, let's say,

> N-Triples. In fact, you can trivially round-trip between them.

All,

Can an "Ontology" be the Subject of a structured description? Yes or No?

In my world view, Description Subjects must be unambiguously identified
via a Name.


Kingsley

>
> Cheers,
> Richard
>
> [1] http://www.w3.org/TR/webarch/#id-resources
>
>
>
>>
>>>
>>> Richard
>>>
>>
>> Ian
>
>

--

Regards,

Kingsley Idehen
President & CEO
OpenLink Software
Web: http://www.openlinksw.com
Weblog: http://www.openlinksw.com/blog/~kidehen
Twitter/Identi.ca: kidehen

[Kingsley Idehen]

All,

Rather than lose ourselves in data representation, let's step back and
rationalize a few things:

1. Information Resource -- Structured Descriptor Document
2. Resource -- Subject of Description (what is being described).


The fact that a Descriptor Document is structured description delivery
mechanism doesn't mean that Documents cannot be themselves description
Subjects. It all about context.

[Ian Davis]

On Fri, May 7, 2010 at 5:53 PM, Richard Cyganiak <ric...@cyganiak.de> wrote:

> To put it another way: An N-Triples serialization of an RDF graph is a
> perfect representation of that graph. The fact that you can round-trip
> between them makes this clear. If it can have a representation, then it's an
> information resource and therefore it can be published as a web document
> (with 200 status code that returns the representation).

Well I argued this way 2 years ago, but it's not the consensus and
it's at odds with e.g. cwm. See messages around
http://lists.w3.org/Archives/Public/www-tag/2008Jan/0071.html

>> Most of the schemas in your second group were authored by me, or
>> by people advised by me, but I now believe they are wrong.
>
> Good to hear that. Any chance of getting these schemas changed, in the
> mid-term?

I'll work on it.

Ian

[Richard Cyganiak]

On 7 May 2010, at 23:20, Stuart A. Yeates wrote:
> On Sat, May 8, 2010 at 9:57 AM, Richard Cyganiak
> <ric...@cyganiak.de> wrote:
>> 2. "http://ex.org/graph" has three representations, one each in RDF/
>> XML,
>> Turtle, and N-Triples.
>
> 2. "http://ex.org/graph" has AT LEAST three representations, ...

Good catch Stuart.

Best,
Richard

[Antoine Zimmermann]

All,


I had a discussion with Richard about that topic, which I want to
summarise here. Richard, please correct me if I missinterpreted the
outcome of the discussion.


2010/5/7 Richard Cyganiak <ric...@cyganiak.de>:

> Antoine,
>
> On 7 May 2010, at 19:03, Antoine Zimmermann wrote:
>>
>> A serialisation of a graph *is not* the graph.
>
> That's correct, but has nothing to do with what I said.
>

[...top of the discussion copied to place things in context...]


The debate was whether an OWL ontology (or an RDF graph) should be
considered as an information resource. I was arguing that ontologies
should be treated as an entity distinct from the information resource
which represents it, while Richard said it is fine to treat ontologies
as information resources.
We finally, agreed that both views are valid. While a graph is not
the same as a file (a byte sequence), it is safe to consider a graph
to be an information resource or even a document. It is safe to say
that the graph *is* the document. This flexibility is in fact needed
to be able to refer to "documents" such as someone's homepage. To
guarantee a little bit of persistence, it must be possible to use a
single URI to refer to the document throughout its lifecycle: the
homepage may change every day, but its URI stay the same. So there is
no problem in having multiple serialisations of the same graph and
still refer to it as a document, all the more so that various formats
can be obtained from the same URI thanks to content negotiation.
Nonetheless, nothing in the Linked Data principles prevents anyone
from assigning a non-information resource URI to a RDF graph or an OWL
ontology. Thereby, it is ok as well to make an explicit distinction
between the ontology as an abstract, non-serialisable concept and the
OWL ontology as a document.

The first approach allows one to use a unique URI for a graph and a
document, which is a bit more convenient.
The second approach may be seen as a little bit more in line with the
OWL DL point of view, which makes a clear separation between what are
ontologies and what are instances.
In OWL DL, something like:

<http://ex.org/onto> a owl:Ontology ;
a foaf:Document .

is not allowed (an ontology cannot be an instance of a DL class).
however, one can write:

<http://ex.org/onto#> a owl:Ontology .
<http://ex.org/onto> a foaf:Document .

Note that this is not necessarily a problem wrt the first approach,
since it is not needed to assert that an information resource is a
foaf:Document. Asserting that an information resource is an
owl:Ontology should be enough.

So all in all, it is more a matter of philosophical taste and I doubt
that one or the other can be strictly sanctioned by the Pedantic Web
group.


Regards,
--
--AZ

[Ian Davis]

On Thu, May 13, 2010 at 2:03 PM, Antoine Zimmermann
<antoine.z...@gmail.com> wrote:

> The debate was whether an OWL ontology (or an RDF graph) should be
> considered as an information resource.  I was arguing that ontologies
> should be treated as an entity distinct from the information resource
> which represents it, while Richard said it is fine to treat ontologies
> as information resources.
> We finally, agreed that both views are valid.  While a graph is not
> the same as a file (a byte sequence), it is safe to consider a graph
> to be an information resource or even a document.  It is safe to say
> that the graph *is* the document.  This flexibility is in fact needed
> to be able to refer to "documents" such as someone's homepage.  To
> guarantee a little bit of persistence, it must be possible to use a
> single URI to refer to the document throughout its lifecycle: the
> homepage may change every day, but its URI stay the same. So there is
> no problem in having multiple serialisations of the same graph and
> still refer to it as a document, all the more so that various formats
> can be obtained from the same URI thanks to content negotiation.

Since I am in the company of pedants, I will ask pedantic questions:

1) What is the truth value of a document? Graphs are either true or false.

2) What is the byte count of an ontology? If something has a byte
count, can it be an ontology?

Ian

[Antoine Zimmermann]

2010/5/13 Ian Davis <m...@iandavis.com>:

What does it mean that a graph is true or false??
I guess documents can also be true or false as well, in some sense.

> 2) What is the byte count of an ontology? If something has a byte
> count, can it be an ontology?

An ontology does not necessarily have a byte count, which does not
prevent it from being a document. In order to have a byte count, a
thing has to be a piece of digital data, which is not the case for all
documents.
If I follow the definition that Richard quoted, an information
resource does not need to be a byte sequence.

--
--AZ

[Ian Davis]

On Thursday, May 13, 2010, Antoine Zimmermann

<antoine.z...@gmail.com> wrote:
> 2010/5/13 Ian Davis <m...@iandavis.com>:
>> On Thu, May 13, 2010 at 2:03 PM, Antoine Zimmermann
>> <antoine.z...@gmail.com> wrote:
>>
>>> The debate was whether an OWL ontology (or an RDF graph) should be
>>> considered as an information resource.  I was arguing that ontologies
>>> should be treated as an entity distinct from the information resource
>>> which represents it, while Richard said it is fine to treat ontologies
>>> as information resources.
>>> We finally, agreed that both views are valid.  While a graph is not
>>> the same as a file (a byte sequence), it is safe to consider a graph
>>> to be an information resource or even a document.  It is safe to say
>>> that the graph *is* the document.  This flexibility is in fact needed
>>> to be able to refer to "documents" such as someone's homepage.  To
>>> guarantee a little bit of persistence, it must be possible to use a
>>> single URI to refer to the document throughout its lifecycle: the
>>> homepage may change every day, but its URI stay the same. So there is
>>> no problem in having multiple serialisations of the same graph and
>>> still refer to it as a document, all the more so that various formats
>>> can be obtained from the same URI thanks to content negotiation.
>>
>>
>> Since I am in the company of pedants, I will ask pedantic questions:
>>
>> 1) What is the truth value of a document? Graphs are either true or false.
>
> What does it mean that a graph is true or false??
> I guess documents can also be true or false as well, in some sense.

The RDF semantics states that a graph (a set of triples) has a truth
value. It's either true or it isn't (eg it is in conflict with itself)

>> 2) What is the byte count of an ontology? If something has a byte
>> count, can it be an ontology?
>
> An ontology does not necessarily have a byte count, which does not
> prevent it from being a document. In order to have a byte count, a
> thing has to be a piece of digital data, which is not the case for all
> documents.
> If I follow the definition that Richard quoted, an information
> resource does not need to be a byte sequence.

The problem is that there is no testable definition of what an
information resource is or isnt. You can check whether someone else
believes something is an information resource if you have a URI for
it: does it return a 200 for an HTTP GET? If yes, then it is an
information resource, if no then the answer is unknown.


>
> --
> --AZ
>

Ian

[Axel Polleres]

On 13 May 2010, at 16:20, Ian Davis wrote:

> On Thu, May 13, 2010 at 2:03 PM, Antoine Zimmermann
> <antoine.z...@gmail.com> wrote:
>
> > The debate was whether an OWL ontology (or an RDF graph) should be
> > considered as an information resource. I was arguing that ontologies
> > should be treated as an entity distinct from the information resource
> > which represents it, while Richard said it is fine to treat ontologies
> > as information resources.
> > We finally, agreed that both views are valid. While a graph is not
> > the same as a file (a byte sequence), it is safe to consider a graph
> > to be an information resource or even a document. It is safe to say
> > that the graph *is* the document. This flexibility is in fact needed
> > to be able to refer to "documents" such as someone's homepage. To
> > guarantee a little bit of persistence, it must be possible to use a
> > single URI to refer to the document throughout its lifecycle: the
> > homepage may change every day, but its URI stay the same. So there is
> > no problem in having multiple serialisations of the same graph and
> > still refer to it as a document, all the more so that various formats
> > can be obtained from the same URI thanks to content negotiation.
>
>
> Since I am in the company of pedants, I will ask pedantic questions:

Ok, you called for pedanticism: ;-)

> 1) What is the truth value of a document? Graphs are either true or false.

But resources aren't. Also, rather the statements in a graph are true or false, not the whole graph per se
(well, yes and no, actually a graph is consistent or inconsistent, which is different from true or false)

> 2) What is the byte count of an ontology? If something has a byte
> count, can it be an ontology?

Not having read the whole thread, admittedly... The correspondence between resources and documents is
a crucial cornerstone of linked data. I wouldn't say that an RDF source is the same as the document
that can be dereferenced at its URI, but I have the expectation to find a document with information about the thing
identified by this URI there. Likewise for an ontology which has a URI, I have the expectation to find a document with
its defining axioms at that URI... this duality is just the last of the Linked data principles, isn't it?

So, of course the ontology isn't the same as the graph, but it is strongly tied to each other,
i.e. I expect that graph to define the ontology and "sloppily" speaking I wouldn't mind if someone
calls that document/graph the ontology.

Will follow up with the guys here at DERI to backtrack the rest of this
discussion to trace what is the actual point of the discussion here. :-)

Axel


> Ian
>

[Richard Cyganiak]

Ian,

On 13 May 2010, at 16:20, Ian Davis wrote:

> Since I am in the company of pedants, I will ask pedantic questions:
>
> 1) What is the truth value of a document? Graphs are either true or
> false.

Most graphs are neither true nor false.

Like many other kinds of documents, RDF graphs are claims about the
arrangement of the world. Unlike many other kinds of documents, RDF
graphs are highly formal claims. Being highly formal makes them easy
to manipulate and transform. For example, it can be possible to
establish wether two graphs claim exactly the same thing, while this
is very hard for other kinds of documents. But this doesn't make it
any easier to evaluate their truth value. Without making a ton of
additional assumptions, the only graphs that have a truth value are
those that are internally contradictory (ex:A owl:differentFrom ex:A)
or trivially true (the empty graph).

> 2) What is the byte count of an ontology? If something has a byte
> count, can it be an ontology?

<foo.owl> can be an RDF/XML-returning information resource and an
ontology at a same time, while still not having a byte count.


I'll make this really easy for you. Here is some terminology:

1. Things that have byte counts = representations
2. Things identified by URIs = resources
3. Things that you GET back from information resources = representations

Note: Thing identified by a URI (resource) != Thing returned when one
GETs the URI (representation).


<foo.owl> is a URI.

<foo.owl> identifies an ontology.

When you do a GET on <foo.owl>, you get back a representation, a <byte
sequence, media type> pair.

The representation does have a byte count.

Having a representation makes <foo.owl> an information resource.

The ontology identified by <foo.owl> does not have a byte count.

> The problem is that there is no testable definition of what an
> information resource is or isnt.

I'm not aware of a testable *necessary* condition for being an
information resource.

But there certainly are testable *sufficient* conditions. If I can
transform something into a <byte sequence, media type> pair, and then
without loss back into exactly the original thing, then it certainly
is an information resource. This is the case with RDF graphs, DOM
trees, database records, pixel images, sound recordings, OWL
ontologies, and other things. All of these are information resources
because there are media types for encoding them.

In other words, proving that something is *not* an information
resource is rather hard (smug pedants can always invoke the
application/x-star-trek-transporter-buffer media type). But for a
limited number of things it can be easily proven that they *are*
information resources. All RDF graphs and all OWL ontologies are among
those things.

Best,
Richard

[1] http://www.w3.org/TR/rdf-mt/#interp

>
> You can check whether someone else
> believes something is an information resource if you have a URI for
> it: does it return a 200 for an HTTP GET? If yes, then it is an
> information resource, if no then the answer is unknown.
>

> Ian

[Ian Davis]

On Thu, May 13, 2010 at 7:29 PM, Richard Cyganiak <ric...@cyganiak.de> wrote:

>
>> 2) What is the byte count of an ontology? If something has a byte
>> count, can it be an ontology?
>
> <foo.owl> can be an RDF/XML-returning information resource and an ontology
> at a same time, while still not having a byte count.
>
>
> I'll make this really easy for you. Here is some terminology:

Richard, you completely miss my point and argue something I've already
pointed out is a position I have taken in the past. It's probably a
better use of your time convincing timbl etc rather than me :)

The point you missed was that I was saying that the document is not
the ontology (it is irrelevant to that point whether they are
information resources). I described two characteristics off the top of
my head that I believe show that they are different things. Just like
maps do not have mountains, documents do not have axioms. Documents
encode axioms, just like maps encode a geography.

Ian

[Thomas Baker]

On Thu, May 13, 2010 at 07:29:17PM +0100, Richard Cyganiak wrote:
> >Since I am in the company of pedants, I will ask pedantic questions:
> >
> >1) What is the truth value of a document? Graphs are either true or
> >false.
>
> Most graphs are neither true nor false.
>
> Like many other kinds of documents, RDF graphs

(Noting that whether an RDF graph is a "kind of document"
seems to be the topic of this thread.)

> are claims about the
> arrangement of the world. Unlike many other kinds of documents, RDF
> graphs are highly formal claims. Being highly formal makes them easy
> to manipulate and transform. For example, it can be possible to
> establish wether two graphs claim exactly the same thing, while this
> is very hard for other kinds of documents. But this doesn't make it
> any easier to evaluate their truth value. Without making a ton of
> additional assumptions, the only graphs that have a truth value are
> those that are internally contradictory (ex:A owl:differentFrom ex:A)
> or trivially true (the empty graph).

Pat Hayes notes that information "such as the exact nature
of the things in the universe - would, regardless of its
intrinsic interest, be irrelevant to the actual truth-values
of any triple" [1]. I take this to mean that determining a
truth-value has strictly to do with logical consistency and
does not try to evaluate whether a graph correctly describes
the actual "arrangement of the world" -- that an RDF graph making
fraudulent or blatantly ridiculous claims could be "true".

> [1] http://www.w3.org/TR/rdf-mt/#interp

> I'll make this really easy for you. Here is some terminology:
>
> 1. Things that have byte counts = representations
> 2. Things identified by URIs = resources
> 3. Things that you GET back from information resources = representations
>
> Note: Thing identified by a URI (resource) != Thing returned when one
> GETs the URI (representation).

Are you saying resources and representations are disjoint?
That a representation is not a resource? That a representation
cannot be identified by a URI?

> Having a representation makes <foo.owl> an information resource.

I thought that _being_ a representation (i.e., of an ontology)
made <foo.owl> an information resource.

I'm not getting the argument...

Tom

--
Tom Baker <tba...@tbaker.de>

[Richard Cyganiak]

Ian,

On 13 May 2010, at 20:31, Ian Davis wrote:
>>> 2) What is the byte count of an ontology? If something has a byte
>>> count, can it be an ontology?
>>
>> <foo.owl> can be an RDF/XML-returning information resource and an
>> ontology
>> at a same time, while still not having a byte count.
>>
>>
>> I'll make this really easy for you. Here is some terminology:
>
> Richard, you completely miss my point and argue something I've already
> pointed out is a position I have taken in the past.

Well, I argue that you can publish a web document that *is* the
ontology, which seems to be in disagreement with your position. If you
agree that the URI </foo.owl> identifies an ontology, then we don't
disagree in substance but only in definitions of words. For me, </
foo.owl> is a web document. I use the terms "web document" and
"information resource" pretty much synonymously.

(The terms "resource" and "representation" have been introduced into
web architecture as precise technical terms to get around the
fuzziness and real-world connotation baggage of words like "web page"
and "web document", but on the Semantic Web "resource" has been
overloaded with the broader meaning of "anything at all", so
"information resource" was introduced to cover the original sense of
"resource", but "information resource" leads to the completely
unhelpful "non-information resource", so I'm avoiding it and using
"web document" instead to mean "thing identified by a URI that has
representations".)

Kingsley will have a field day with this.

Best,
Richard

[Richard Cyganiak]

On 13 May 2010, at 21:02, Thomas Baker wrote:
> (Noting that whether an RDF graph is a "kind of document"
> seems to be the topic of this thread.)

Which of course depends on your definition of document -- see my
response to Ian.

>> 1. Things that have byte counts = representations
>> 2. Things identified by URIs = resources
>> 3. Things that you GET back from information resources =
>> representations
>>
>> Note: Thing identified by a URI (resource) != Thing returned when one
>> GETs the URI (representation).
>
> Are you saying resources and representations are disjoint?

I meant to stress that they are two different concepts. This doesn't
mean they are necessarily disjoint.

> That a representation is not a resource? That a representation
> cannot be identified by a URI?

Well, I can mint a URI for *anything*.

But the association between URI and referent will only live in my head
unless I inform the world about it. And the webby way of doing that is
to set up a web server that responds to HTTP requests to the URI with
useful information about it. Doing that for a URI that identifies a
representation is possible, and sometimes useful, but a corner case.

The definition of "representation" in web architecture is: "a byte
sequence with a media type (and perhaps some other metadata)". This is
not the same meaning as "representation" in philosophy, as Pat Hayes
has complained many times.

The main role of representations in web architecture is: They are the
things that are being sent over the wire in HTTP messages. So when you
do a GET on the URI <foo.owl> which identifies some resource, then a
*representation of that resource* is being sent over the wire. It is
*not* the resource that's being sent over the wire; the resource stays
wherever it was. That's the big distinction between resources and
representations in webarch. The other big distinction is that
resources can change over time, while representations are immutable
per definition. So if I edit the ontology, then tomorrow the URI
<foo.owl> will still identify the same *resource*, but a GET request
would return a different *representation*.

>> Having a representation makes <foo.owl> an information resource.
>
> I thought that _being_ a representation (i.e., of an ontology)
> made <foo.owl> an information resource.

My statement was a somewhat sloppy restatement of the httpRange-14
decision. Slightly more precise: "Responding to HTTP GET with a
representation (this implies a 200 OK status code) makes <foo.owl> an
information resource."

Just to tease apart the entire chain of thought: <http://example.com/foo.owl
> is a URI (per definition of "URI"). It identifies a resource (per
definition of "URI"). If I do an HTTP GET on the URI and the response
status code is 200, then the payload sent in the body of the 200
response is a representation of that resource (per definition of
"representation"). So, by doing the HTTP GET request, we found out
that the resource "has" a representation (it's enclosed in the
response message). We also found out that it identifies an information
resource (because of the 200 code, per httpRange-14).

Best,
Richard

[Thomas Baker]

On Fri, May 14, 2010 at 12:31:00AM +0100, Richard Cyganiak wrote:
> Well, I argue that you can publish a web document that *is* the
> ontology, which seems to be in disagreement with your position. If you
> agree that the URI </foo.owl> identifies an ontology, then we don't
> disagree in substance but only in definitions of words. For me, </
> foo.owl> is a web document. I use the terms "web document" and
> "information resource" pretty much synonymously.

Sorry - just to rule out one possible source of confusion...

Are we assuming that </foo.owl> is the name of a
file/web-document/information-resource on a server somewhere?
I understand you to say that if it is, it _could_ be considered
the ontology.

In principle, might </foo.owl> not also be
an non-information resource represented by a
file/web-document/information-resource named, say,
</foo-owl-20100107.rdf>? Would this latter </foo.owl> be an
ontology? Would </foo-owl-20100107.rdf> be an ontology?
Both?

[Andrae Muys]

1. Documents don't have truth values - Documents are Representations
of Resources that themselves may support a monomorphism onto a logic
for which your question may or may not make sense. Think
Representamen-Interpretant-Object.

In this case we have an RDF-serialization of an OWL-ontology, which
itself is an RDF-graph with a mapping to FoL equal to the conjunction
of the individual triples <S,P,O> to the propositions P(S,O). Rinse
and repeat a few times as you pass though denotational OWL semantics,
and ultimately the connotational semantics of the ontology's domain,
and you will start to approach an expression that might have a value
of 'true' or 'false'.

Your mistake here is to fail to realise RDF graphs are
epistemological. The don't have truth values - they are instead
assertions of belief, assertions that are (in isolation) only
consistent, inconsistent, or undecidable (for some reason).

2. Pedantically, ontologies don't have byte-counts, and nothing with a
byte-count can be an ontology. Pragmatically, as there is a
monomorphism from the representation to ontology, there is often no
need to make the distinction in practice.

Andrae

--
Andrae Muys,
Senior Software Engineer, Cambia
Initiative for Open Innovation (IOI)
Cambia@QUT, G301, 2 George Street, Brisbane Qld 4000, Australia
+61 414 517 882 (mobile) +61 7 3138 4538 (work) +61 7 3138 4405 (fax)

[Richard Cyganiak]

Hi Thomas,

On 14 May 2010, at 04:34, Thomas Baker wrote:

> On Fri, May 14, 2010 at 12:31:00AM +0100, Richard Cyganiak wrote:
>> Well, I argue that you can publish a web document that *is* the
>> ontology, which seems to be in disagreement with your position. If
>> you
>> agree that the URI </foo.owl> identifies an ontology, then we don't
>> disagree in substance but only in definitions of words. For me, </
>> foo.owl> is a web document. I use the terms "web document" and
>> "information resource" pretty much synonymously.
>
> Sorry - just to rule out one possible source of confusion...
>
> Are we assuming that </foo.owl> is the name of a
> file/web-document/information-resource on a server somewhere?

That doesn't really matter. The only thing that matters is what
clients can learn by prodding the resource via the HTTP protocol. How
the server obtains the byte sequences that it might send in response
-- wether they come from a file in the server's file system, wether
they are generated dynamically from execution of a PHP script or Java
servlet, wether an army of interns types them in real time -- is an
implementation detail. The relevant specs (HTTP, URIs, RDF, etc)
define an interface for communication between systems, and not an
implementation.

> I understand you to say that if it is, it _could_ be considered
> the ontology.

I say that a 200 response at </foo.owl> with for example an RDF/XML
serialization is consistent with the claim that </foo.owl> is an
ontology (and hence, it's ok to say that "</foo.owl> a owl:Ontology"
inside the RDF/XML).

> In principle, might </foo.owl> not also be
> an non-information resource represented by a

I assume you meant to say "described by ..." here? Remember, something
that has a representation is, by definition, an information resource!

"Non-information resource" is an unfortunate term and I recommend
avoiding it. Just say "resource", or "resource without any
representations".

> file/web-document/information-resource named, say, </foo-

> owl-20100107.rdf>? Would this latter </foo.owl> be an ontology?
> Would </foo-owl-20100107.rdf> be an ontology?
> Both?

Well it's up to the URI owner to tell us what the URIs identify, and
from the outside this is usually impossible to tell unless there is an
explicit statement from the URI owner. But if the URI owner told us
that either or both of these URIs are ontologies, then that would be
fine, and I see no contradiction.

Hope that helps,
Richard