India is Class and Organization?

[aharth]

Hi,

any reason why http://umbel.org/umbel/sc/India.rdf says

<http://umbel.org/umbel/sc/India> a <http://www.w3.org/2002/07/
owl#Class>; a
<http://umbel.org/umbel/sc/Organization"/> .

?

Regards,
Andreas.

[Frederick Giasson]

Hi Andreas,

> any reason why http://umbel.org/umbel/sc/India.rdf says
>
> <http://umbel.org/umbel/sc/India> a <http://www.w3.org/2002/07/
> owl#Class>; a
> <http://umbel.org/umbel/sc/Organization"/> .
>

We explicit this since umbel:India is both a class and an instance of a
class (so an individual). Please refer to this document for an "in-deep"
explanation of what is happening here:

http://umbel.org/doc/UMBELOntology_vA1.pdf

Hope this helps.

Thanks,

Take care,

Fred

[Blake Shepard]

Fred,  I think we have discussed this before, but I still have some lingering questions about these matters which seem to me to indicate a fundamental difference between the Umbel ontology and the Cyc ontology that might be worth making explicit.  I'll characterize the relevant aspects of the Cyc ontology in detail, and then raise some questions about pertinent aspects of the Umbel ontology.

In the Cyc ontology, the collection of all collections (#$Collection) is disjoint with the collection of all individuals (#$Individual).  Indeed, in the Cyc ontology, #$Individual and #$Collection impose a partition on the collection of all things: each instance of any collection in the Cyc ontology is either an individual or a collection, but is never both.  For example, in Cyc #$India is an instance of the collection #$Country (which is a subcollection of #$Individual), and the collection #$Country is an instance of the collection #$Thing. Thus, in Cyc, #$India is an individual but not a collection and #$Country is a collection but not an individual, even though both #$India and #$Country are instances of other collections. 

Other relevant features of collections and individuals in the in the Cyc ontology are: only collections can have instances, and only individuals can exist in space. 

I think that if an ontology that has the following features, everything in the ontology will be considered to be an individual:

1. The ontology has a "top node" class of which everything is an instance, e.g. something like the the Cyc collection #$Thing.

2. The ontology is structured according to the principle that being an instance of any class is sufficient for being an individual.

Feature 2 does not characterize the Cyc ontology, so relative to the Cyc ontology "individual" and "instance" are orthogonal notions.

Do features 1 and 2 characterize the Umbel ontology?

I have studied the Umbel documentation, but I remain unsure of why India is characterized as a class in the Umbel ontology.  In my mind, classes must be able to have instances, but I have been unsuccessful in my struggle to understand what could be an instance of India.  

What in the Umbel ontology might be an instance of India?  

- Blake

[Mike Bergman]

Hi Blake,

Let me take a crack at this, which may not be intellectually
rigorous or satisfying, but we hope is pragmatic ...

These are great questions and reflect, I think, the different
purposes and world views between Cyc and UMBEL. I wrote on this
in general earlier in 'Stuck in the Middle with You ...' [1].

UMBEL's purpose is to be a gathering point for Web content and
datasets that are "about" similar subjects or concepts. Our
role, as I see it, is merely as a traffic cop to help content
find similar content.

Every subject concept in UMBEL *is* a class with external
instances (Web content and datasets) related to it by the isAbout
predicate [2, actually, looking over this vocabulary suggests to
me the importance to improve our definition to refer to instances].

We realize that this "gluing" or "reference" function has little
reasoning or reference power. That is one reason why we have
tried to maintain a concept correspondence with Cyc: those who
need the rigor of a self-consistent reasoning framework can use
UMBEL as the doorway to your framework and services.

As for India, we made a bit of an arbitrary decision to include
both countries and their provinces or states as "subject
concepts" in UMBEL because we felt that appropriate to capture
the granularity and scope elsewhere in UMBEL. So, for us, the
question was not whether there were instances internally in UMBEL
for the class India, but whether there was a useful purpose to
the external world to say that their content "was about" (an
instance of that class) India.

As you know, we have had similar discussions with Cycorp around
the whole area of academic subjects, which you have defined as
FieldOfStudy. For UMBEL's purposes it would be helpful to enable
outside parties to say their content was "about" biology or
astronomy, even though the strict internal treatment (as you
note, Individuals v Collections) in Cyc does not easily allow this.

I think some real progress has been made there in finding a good
compromise (actually, "interface") between UMBEL and Cyc in terms
of how Cyc predicates can be mapped to our SKOS equivalents.
Maybe this India example points to another area where we should
sharpen our pencils a bit.

BTW, if you have other craw stickers, spit 'em out!

Mike

[1] http://www.mkbergman.com/?p=441
[2] http://www.umbel.org/technical_documentation.html#vocabulary

>
> - Blake
>
>
> On Wed, Feb 11, 2009 at 8:26 PM, Frederick Giasson <fr...@fgiasson.com
> <mailto:fr...@fgiasson.com>> wrote:
>
>
> Hi Andreas,
>
> > any reason why http://umbel.org/umbel/sc/India.rdf says
> >
> > <http://umbel.org/umbel/sc/India> a <http://www.w3.org/2002/07/
> > owl#Class>; a
> > <http://umbel.org/umbel/sc/Organization"/> .
> >
> We explicit this since umbel:India is both a class and an instance of a
> class (so an individual). Please refer to this document for an "in-deep"
> explanation of what is happening here:
>
> http://umbel.org/doc/UMBELOntology_vA1.pdf
>
>
> Hope this helps.
>
>
> Thanks,
>
>
> Take care,
>
>
>
> Fred
>
>
>
>
> >

--
__________________________________________

Michael K. Bergman
CEO Structured Dynamics LLC
319.621.5225
skype:michaelkbergman
http://structureddynamics.com
http://mkbergman.com
http://www.linkedin.com/in/mkbergman
__________________________________________

[Frederick Giasson]

Hi Blake,

Well, thanks for bringing up this good old discussion :)

First, I don't think owl:Class == cyc:Collection.

But if I remember our discussion at cycorp, here was the problem we
found (not sure if it was with you or Micheal):

Prior our meeting, we were using owl:equivalentClass to link any subject
concept to any Cyc collection. However, we found a bug in this mapping:
some of the Cyc resources we had, were not collections, but individuals.
So, yes, you were right, there was an issue in using owl:equivalentClass
to make that linkage.

When I come back home, I fixed this situation so that we use the
owl:equivalentClass to linka SC to a cyc:Collection (which are classes
in RDF representation of OpenCyc); and to use owl:sameAs to link the
umbel:SubjectConcepts that comes from cyc:Individual. This is this
distinction between "owl:equivalentClass" and "owl:sameAs" that explicit
the equivalence of a class (equivalentClass) and equivalence of an
individual (sameAs).

So, if you take a look at this page:

http://tinyurl.com/btqmuu

You will notice that umbel:India sameAs cyc:India (and not equivalentClass).

Does this help putting some light on what is happening here?

Thanks!

Take care,

Fred

[Jack Park]

FWIW, there was a short thread in the Ontolog List starting at [1]
where Chris Partridge asked "what 'Individual' in OWL meant?"

The thread appears to continue at [2]. Wikipedia [3] had this to say:
"Concept as abstracted knowledge is explicitly implemented by
individuals and classes in the geopolitical ontology. Individual is
defined as an object perceived from the real world in geopolitical
domain (i.e. "Ethiopia", and "least developed countries").

Class is defined as a set of individuals sharing their common properties."

Wandering along a garden path using Wikipedia's comments, we imagine a
*concept* that represents what a shoe is, its "shoeness". We then
imagine a *class* of shoes, say, tennis shoes, then we imagine
*individuals* that are those, say, Nike tennis shoes you might be
wearing right now.

I wonder where the term "subject concept" fits into that scheme.

BTW: Chris Partridge has a book out _Business Objects: Re-Engineering
for Re-Use_ which appears to be a fairly in-depth practical knowledge
engineering cookbook. Not sure; my copy of it just now arrived.

Jack
[1] http://ontolog.cim3.net/forum/ontolog-forum/2009-02/msg00256.html
[2]http://ontolog.cim3.net/forum/ontolog-forum/2009-02/msg00314.html
[3] http://en.wikipedia.org/wiki/Geopolitical_ontology

[aharth]

Hi,

I agree with the comments questioning why India should be of type
owl:Class.

I've read the manual and still don't get why subject concepts are both
classes and individuals, but hey that's your modeling choice.

What worries me is that questionable UMBEL statements pollute other
datasets via chains of inferences.

E.g. from the UMBEL data you can arrive at the fact

<http://www.w3.org/2006/03/wn/wn20/instances/synset-India-noun-1> a
foaf:Organization .

which I strongly disagree with. In my dictionary, India is a country,
and not an organization.

$ dict organization | grep -e country
$
$ dict country | grep -e organization
$

Just to illustrate how the infection spreads:

http://umbel.org/umbel/sc/India.rdf says:

umbel:India a umbel:Organization .
umbel:India = <http://www.w3.org/2006/03/wn/wn20/instances/synset-
India-noun-1> .

http://umbel.org/umbel/sc/Organization.rdf says:

umbel:Organization owl:equivalentClass foaf:Organization .

which leads to the statement I disagree with:

<http://www.w3.org/2006/03/wn/wn20/instances/synset-India-noun-1> a
foaf:Organization .

It also leads to:

<http://www.w3.org/2006/03/wn/wn20/instances/synset-India-noun-1> a
owl:Class .

which is equally questionable.

I guess it would make sense to drop the owl:sameAs and
owl:equivalentClass statements in UMBEL which *do not* link to
equivalent things, to stop polluting external datasets.

Regards,
Andreas.

[Blake Shepard]

Based on what you say, Mike, my interest increases in how UMBEL accomodates the relations "is an instance of" and "is about". 

Does UMBEL only allow classes to be about other things?  If so, that seems odd to me since I believe that uin the actal world individuals can be about other things.  Fodor's guide to India, for example, is an individual that is about India.  Similarly, biology is an individual field of study that is about living organisms.

Fred, thank you also for your input.  I appreciate that the loose mapping between Cyc and UMBEL may alleviate the tension between the Individual/Collection disjointness in Cyc and the fact that some entities in UMBEL are both individuals and classes.

However, I do worry that there is a tension between UMBEL's notion that India is a class and the actual structure of the world we live in.  More generally, I worry that UMBEL is structured around the idea that only classes can be about other things, whereas I do not believe that is true of our actual world.  I therefore have a lingering belief that UMBEL would be better off allowing for non-class "subject concepts".

- Blake

[Frederick Giasson]

Hi Andreas!

> I agree with the comments questioning why India should be of type
> owl:Class.
>
> I've read the manual and still don't get why subject concepts are both
> classes and individuals, but hey that's your modeling choice.
>

The premise here, and to everything related to umbel, is that umbel is
OWL Full 1.1

We will try to make it OWL DL (some profile; don't know which yet) 2.0
as soon as the standard is accepted; but in mean time, everything is OWL
Full 1.1; and this is quite important for the following.

> What worries me is that questionable UMBEL statements pollute other
> datasets via chains of inferences.
>

Well please take care here with the word "pollute". First of all, I tend
to find that more and more people think that what is "in the semantic
web" is all right. In fact 95% of the Web is crap; and I would guess
that 99% of the semantic web is crap too. So the first thing people
should tell themselves is: let consider that everything out there is
crap; and from there, try to put some order in that so that we find some
gold mine.

The point here is that you have to reason over what you have, and you
have to make sure that things are consistent by themselves. Don't see
the "semantic web" as something that is in the universe like dark
matter; and try to create your own environment with its own laws and way
to see and *reason* over things.

Then you will be able to leverage the "semantic web" and do wonderful
things.

If you read the whole UMBEL technical document, you probably noticed
that we talked about "UMBEL's World View". This mean that UMBEL tries to
be consistent in itself, and consistently link to external ontologies. I
am not saying that it is the case, but the work we have done so far on
the structure tell us that it "holds" together so far.

UMBEL's *doesn't* pollute any other dataset except if you trust linkage
you do that create reasoning problems.

> E.g. from the UMBEL data you can arrive at the fact
>
> <http://www.w3.org/2006/03/wn/wn20/instances/synset-India-noun-1> a
> foaf:Organization .
>
> which I strongly disagree with. In my dictionary, India is a country,
> and not an organization.
>

Well; everything depends on how you see a country. Is it a political
entity? A defined set of land by convention? By Laws? Is a country
something defined by its culture(s) and inhabitants?

In my dictionary, a country can be seen as many things. However, what is
important to build great stuff on the semantic web: is to be consistent
in itself. Otherwise you are doomed ;)

This is why, I think, the concept of "World views" is important.
Otherwise you will always see such issues, even ion 25 000 years. I see
the world with my eyes and experience; you see the world with your eyes
and experience; why would it be different for these systems?

This is a metaphor; but an important one.

> $ dict organization | grep -e country
> $
> $ dict country | grep -e organization
> $
>
> Just to illustrate how the infection spreads:
>
> http://umbel.org/umbel/sc/India.rdf says:
>
> umbel:India a umbel:Organization .
>

Yes, this triple exist.

> umbel:India = <http://www.w3.org/2006/03/wn/wn20/instances/synset-
> India-noun-1> .
>

No this triple doesn't exist in UMBEL's World View. (it is not defined
in what we published)

Is this a linkage you created by yourself, or a linkage you found
somewhere "in the semantic web"?

This is exactly what I was talking about above: if you trust assertions
that haven't been validated; or wrongly validated, then yes, weird
assertions can be created, and such things could happen.

> http://umbel.org/umbel/sc/Organization.rdf says:
>
> umbel:Organization owl:equivalentClass foaf:Organization .
>

Exactly. This is part of the linkage between umbel and external
ontologies such as FOAF.

> which leads to the statement I disagree with:
>
> <http://www.w3.org/2006/03/wn/wn20/instances/synset-India-noun-1> a
> foaf:Organization .
>

Given the assertion above, that *is not* from UMBEL, that assertion is
right.

> It also leads to:
>
> <http://www.w3.org/2006/03/wn/wn20/instances/synset-India-noun-1> a
> owl:Class .
>

No. synset-India-noun-1 is a member of owl:Thing. If you want an
individual of a class being a class itself you have to instantiate it
has XYZ a owl:Class. To define a class, you have no choice, you have to
assert rdf:type rdfs:Class or owl:Class; or you you can do it using
rdfs:subClassOf (in that case, you infert rdf:type Class out of the
subClassOf property).

> which is equally questionable.
>
I hope; particularly considering the above ;)

> I guess it would make sense to drop the owl:sameAs and
> owl:equivalentClass statements in UMBEL which *do not* link to
> equivalent things, to stop polluting external datasets.
>

Please take care when you do such assertions.

UMBEL don't *pollute* anything. You use umbel; you load the external
linkage; and you create new linkage. If the result is a mess because of
some wrong assertions have been created, then you have to fix it. UMBEL
doesn't "pollute" anything out there. It is what you will do with it
that will determine if you pollute something; and if it pollute
something, then it only pollute the system that uses it; and the "other
datasets out there".

Thanks for this discussion, you raised important points that have to be
discussed and understood.

Take care!

Fred

[aharth]

Hi,

On Feb 13, 2:44 am, Frederick Giasson <f...@fgiasson.com> wrote:
> > umbel:India = <http://www.w3.org/2006/03/wn/wn20/instances/synset-India-noun-1> .

>
> No this triple doesn't exist in UMBEL's World View. (it is not defined
> in what we published)
>
> Is this a linkage you created by yourself, or a linkage you found
> somewhere "in the semantic web"?

yes you're right, the inference is done via OpenCyc:

umbel:India = opencyc:India = <http://www.w3.org/2006/03/wn/wn20/
instances/synset-India-noun-1> .

which leads to the statements which don't fit my world view.

> > I guess it would make sense to drop the owl:sameAs and
> > owl:equivalentClass statements in UMBEL which *do not* link to
> > equivalent things, to stop polluting external datasets.
>
> Please take care when you do such assertions.

Hm ok. I guess I'll exclude UMBEL from my crawls then to avoid the
inferences
for which there is no concensus. In general, before I take this
drastic
measure I try to get the things fixed at the source, which I hope
benefits other
data consumers as well.

> Thanks for this discussion, you raised important points that have to be
> discussed and  understood.

Yep, my pleasure!

Regards,
Andreas.

[Frederick Giasson]

Hi Andreas!

>>> umbel:India = <http://www.w3.org/2006/03/wn/wn20/instances/synset-India-noun-1> .
>>>
>> No this triple doesn't exist in UMBEL's World View. (it is not defined
>> in what we published)
>>
>> Is this a linkage you created by yourself, or a linkage you found
>> somewhere "in the semantic web"?
>>
>
> yes you're right, the inference is done via OpenCyc:
>
> umbel:India = opencyc:India = <http://www.w3.org/2006/03/wn/wn20/
> instances/synset-India-noun-1> .
>
> which leads to the statements which don't fit my world view.
>
>
>>> I guess it would make sense to drop the owl:sameAs and
>>> owl:equivalentClass statements in UMBEL which *do not* link to
>>> equivalent things, to stop polluting external datasets.
>>>
>> Please take care when you do such assertions.
>>
>
> Hm ok. I guess I'll exclude UMBEL from my crawls then to avoid the
> inferences
> for which there is no concensus. In general, before I take this
> drastic
> measure I try to get the things fixed at the source, which I hope
> benefits other
> data consumers as well.
>

Well; it is up to you to use umbel or not. However you raised an
interesting point with this wordnet entity.

There is still a problem with that I think, considering UMBEL or not. I
shared this concern with Cyc, however I will resume it here.

The problem here is the use of owl:sameAs. Saying that the individual
opencyc:India owl:sameAs wn:India entails that these two individuals are
the same in all possible sense (they are identical). This is even much
more powerful a relationship than owl:equivalentClass for example.

So, asserting this fact, means that opencyc:India and wn:India becomes
identical, so that opencyc:India becomes a NounSynsetand that wn:India
becomes an opencyc:IndependantCountry for example. Which, I agree, is
questionnable.

However this entailments doesn't come from UMBEL, but the linkage
between OpenCyc and WN.

Considering that Wordnet's purpose is to describe words and their
relationship, I suggested to use another property to link opencyc
individuals and classes to wordnet individuals. Something such as
rdfs:seeAlso, or to create a subProperty of rdfs:seeAlso that would be
more specific for this purpose.

But the same discussion we had in our earlier mails apply to this case:
people have to check if linkage works in their "world view"; so if it is
consistent with their minding; and use or reject the linkages appropriately.

Does this make sense?

Thanks!

Take care,

Fred