Duane,
Just on one point:
Ø [DN] WordNet is
where I was also thinking of going but I find it incomplete. For example,
take the definitions of river and creek:
[Rive] 1. a large natural stream of water (larger than a creek); "the
river was navigable for 50 miles"
wordnet.princeton.edu/perl/webwn
Ø [creek] 2. a
natural stream of water smaller than a river (and often a tributary of a river)
Circular reasoning or no loose ends?
Actually, there’s nothing wrong with cross-reference, if
it proves necessary – it happens a lot at the basic levels. What it
means is that whatever module (context) holds one concept must also hold the
other – they are interdependent (like ‘metric space’ and ‘dimension’
). In this case, since there are no objective criteria (e.g. ‘a river
must be at least 50 feet across at some point in its length when it is not at
flood stage’ or ‘ a river must be at least 3 feet deep for one mile
from its mouth’), then that signals that the definitions may be
context-dependent (e.g. different locations will have different criteria,
different users will use different labels). Then it may be necessary
(depending on the application) at some point to introduce more precise
definitions depending on the context, or just refer to the act of affixing a
label. I recall one example of a government study of funding in chemistry
departments where the only criterion they could find for whether some part of a
university is or is not a ‘Chemistry Department’ was whether the
university called it a “Chemistry Department”. This is not
actually circular, because the label affixed and the object itself are not the
same thing.
But as I suggested, I think that one needs to try as hard as one
can to get agreement within some large community to use the same basic
ontology, and only give up and look for alternatives after reasonable efforts
have failed.
As for WordNet, it is an excellent start, but needs to be formalized,
which means reorganized, especially at the top levels. That might serve as one
of the main resources for a tagger’s ontology.
Pat
Patrick Cassidy
MICRA, Inc.
908-561-3416
cell: 908-565-4053
cassidy@xxxxxxxxx
From:
ontolog-forum-bounces@xxxxxxxxxxxxxxxx [mailto:ontolog-forum-bounces@xxxxxxxxxxxxxxxx]
On Behalf Of Duane Nickull
Sent: Thursday, April 23, 2009 12:40 PM
To: [ontolog-forum]
Subject: Re: [ontolog-forum] Namespaces for ontologies
Pat:
Major thanks for this. As for the caveat, I had similar thoughts. What
I want to do is a limited roll out and only those who are willing to initially
learn the system would be allowed to make the references until it is clear that
it works or not. This probably sounds very elitist but that is how I want
to proceed.
The time issue is also something I have thought about however the goal is to
put the theory into practice and see if it is or is not useful. As
with complexity, I would assume that the real world results will speak for
themselves.
Wordnet is where I was also thinking of going but I find it incomplete.
For example, take the definitions of river and creek:
1. a large natural stream of water (larger than a creek); "the river was
navigable for 50 miles"
wordnet.princeton.edu/perl/webwn
2. a natural stream of water smaller than a river (and often a tributary of a
river)
Circular reasoning or no loose ends?
I guess I am curious to see this ported to the masses to see where it fails or
works.
Duane
On 4/22/09 5:45 PM, "Patrick Cassidy" <pat@xxxxxxxxx>
wrote:
Duane –
I don’t have any suggestions about how one might best use
URL’s to reference ontology elements. However, I do have some
thought about whether one might or might not want to.
I’m not quite sure what you intend to accomplish by allowing
taggers to link their tags to upper ontology elements from any old ontology.
Since the relationships of the upper ontologies to each other have not
yet been well developed, it would only be useful for those who have chosen to
tag by reference to the same ontology. That might be somewhat useful, but
see the caveat below.
If, however, you think this is a way to bring the power of crowds to
actually finding the relations among ontologies (i.e. a tag referenced to more
than one ontology creates a putative identity link), then this is an
interesting new way to go about the ontology mapping task. But that
method has a problem:
Caveat: since upper ontologies are hard to understand, I suspect few taggers
will have the patience to work though those structures except to find the most
common tags like “person” or “organization”, and in
those cases the bare tag would probably have a common-sense meaning which
others would understand without the ontology reference.
An example of the problem is the axiom you gave, apparently taken from SUMO.
To interpret it, one needs to read the documentation for the
‘time’ relation. For perspicuity, I would recommend
that any project of that type at least rename the relations with a verbal form
– in this case ‘existedAtTime’. That can help users at
least avoid having to puzzle over their recollection of the meaning and
polarity of the relation. The whole relation means that an Object exists
at every time point within *some* time interval. I have never been able
to figure out under what circumstances this axiom might be useful in inference,
though it makes a sensible philosophical statement about physical objects.
It would be good to see crowds adopt a single ontology that is well-structured
but easy to understand. In that case I think the best tactic is to have
such an ontology developed by the taggers themselves for their purpose, with
the help of some willing ontologists. The existing ontologies can be
taken as starting points, but having the taggers use a common ontology would be
a lot more effective.
If you think that there is a clamor among taggers to use some particular
existing ontology, so that that one of them must be included in the mix that
the taggers use, I would be very interested to know which ones have that
distinction. So far the biggest public tagging corpora I know of
reference WordNet.
As for versioning, in the cases I am familiar with, new
elements tend to get added, but the basic structure of the main elements seldom
changes (though the WordNet indexes change). There are occasional
exceptions, one of which happened with the way Cyc handles attributes, a few
years ago. But since the process you describe is so subject
to variation that I think that the variation added by ontology versions will
not be significant, and I wouldn’t worry about versioning – at
least beyond the 1.0 stage.
Pat
Patrick Cassidy
MICRA, Inc.
908-561-3416
cell: 908-565-4053
cassidy@xxxxxxxxx
From: ontolog-forum-bounces@xxxxxxxxxxxxxxxx
[mailto:ontolog-forum-bounces@xxxxxxxxxxxxxxxx]
On Behalf Of Duane Nickull
Sent: Wednesday, April 22, 2009 6:10 PM
To: [ontolog-forum]
Subject: [ontolog-forum] Namespaces for ontologies
Hey all:
I am right now placing FOL binary relationships into the associations binding
for labeled relations in a test Registry-Repository. I am doing this
because I got sick of just talking about this and not actually coding and doing
something to solve the problem so the Canadian cowboy instinct to code first
and see where problems arise has taken over.
I have run into a problem that I would appreciate input on. My goal is to bind
something like this:
(=>
(instance ?OBJ Object)
(exists
(?TIME1
?TIME2)
(and
(instance
?TIME1 TimePoint)
(instance
?TIME2 TimePoint)
(before
?TIME1 ?TIME2)
(forall
(?TIME)
(=>
(and
(beforeOrEqual
?TIME1 ?TIME)
(beforeOrEqual
?TIME ?TIME2))
(time
?OBJ ?TIME))))))
..to a registry-repository instance to allow folksonomy tags to reference the
upper level ontology classes the folksonomy tag owners believe they belong to.
Additionally, each instance of a folksonomy tag may have * relationships
to other ontology classes or even other folksonomy tags. The latter
relastionships can be defined in terms of contrained relationship tags like
“synonym, disjoint, etc.”.
I want to represent all upper ontologies however some of them contain subtle
nuances between their terms. Dolce, SUMO and others have defined binary
relationships like transitive, intransitive, reflexive, irreflexive,
symmetrical as well as some partial ontologies. The problem is that there
are no namespace qualifications for these so I want to introduce that into my
work. I was planning on just using the root URL’s for each work
however there are versions possible in some of the work.
I would like this to be in the firm of
<upper_ontology_identifier>+<version_or_instance>+<uuid>
as a classifier followed by the term label such as “transitive”.
I will probably use URI’s for the UUID.
Question:
Has anyone ever come across a similar problem and if so, how did they solve it?
Thoughts and comments welcome too.
Duane