Thread Links Date Links
Thread Prev Thread Next Thread Index Date Prev Date Next Date Index

SUO: Re: Industry takeover




John,

At 06:39 AM 4/29/2003 -0400, John F. Sowa wrote:
>Jean-Luc, Adam, and Robert,
>
>The announcement that Google bought Applied
>Semantics means that a 600-pound gorilla has
>jumped into our little ontology wading pool:
>
>JLD> Too bad for you guys...
>>
>>Big money is going to tell you what a
>>"Standard Upper Ontology" is:
>
>http://www.appliedsemantics.com/ne/ne_pr_042303.html
>http://www.appliedsemantics.com/company/company_technology.html
>http://www.appliedsemantics.com/ne/ne_news_070102.html
>
>>No less monolithic, orwellian and conservative
>>than you could ever dream of!  We all can rest now
>>and just buy it.
>
>Some observations:
>
>1. The white papers on the Applied Semantics web site
>    make it clear that they are not doing reasoning,
>    expert systems, or anything remotely resembling
>    what Cyc is doing.
>
>2. What they call an ontology is what I called a
>    "terminological ontology" in the glossary on my
>    web site:
>
>    http://www.jfsowa.com/ontology/gloss.htm
><http://www.jfsowa.com/ontology/gloss.htm>
>    Therminological ontologies are important starting
>    points for more detailed axiomatizations.  But they
>    are not doing what the SUO is attempting to do.
>
>3. Nevertheless, you can't ignore a big gorilla in
>    your wading pool.  Even if its intentions are benign,
>    it can crush you in a playful embrace.  So you must
>    pay attention and make sure that you treat it with
>    all due respect.

I agree.  It sounds like the sort of text processing that many other 
companies do, along with the sort of taxonomy that appears in Yahoo.

>This brings us to something that Adam Pease mentioned
>in an earlier note about IFF and its relevance to SUO:
>
>AP> Our charter doesn't address ontology mapping or
>>translation.  IFF seems like a product that might help
>>one compose ontologies in order to create a proposed
>>standard for the SUO, but that it is not a relevant
>>proposal itself.
>>
>>In terms of John's proposal for a combined effort,
>>I could similarly see IFF being used in combining SUMO,
>>Cyc, or other ontologies, but the resulting artifact
>>would be the proposal to SUO, not the original
>>ontologies or IFF.
>
>The important point is that no ontology can ever be
>a static, stand-alone artifact.  The terminologies
>used by Applied Sciences are overwhelmingly larger
>than Cyc in terms of the number of concept types
>(or word senses).  Somehow, the SUO will have to
>be related to such things, and the methods for
>composing or relating ontologies must be part of
>any practical methodology for using ontologies.
>
>Whether or not the methodologies for using,
>relating, or composing ontologies are in the SUO
>charter or are implied by clauses that are in the
>charter is not clear.  But if we don't address
>those issues, we are going to be crushed by
>the gorilla.

I agree, but that doesn't mean that's what this group is supposed to be 
focusing on.  Often when standards processes progress, new needs are 
discovered, and new groups spin off.



>AP> An additional concern that I have is that
>>the utility, as opposed to the mathematical
>>possibility, of employing IFF to that end is
>>also unsupported by any examples or practical
>>experience, at least so far as messages to this
>>mailing list are concerned.  There are many folks
>>working on ontology mapping.  What evidence is
>>there that a category theory approach allows one
>>to merge ontologies faster, or more easily than
>>other approaches, including that of emacs and
>>a human ontologist with no other tools?
>
>My brief justification:  IFF is a structured
>framework for using logic to do theory revision,
>editing, merging, and comparison while ensuring
>that the results are consistent.

The purported use is not in question, but a paragraph of explanation does 
not comprise evidence of utility.

>The methods for "aligning" ontologies, which Ed Hovy
>and others have discussed, have only addressed the
>problem of relating the terminology.  They have not
>gone to the deeper level of checking whether the
>associated axioms are consistent.  That is the most
>critical part of the problem, and it cannot be
>solved without some serious logic-based tools.

I agree.  A number of different DARPA contractors are working in ontology 
mapping.  Many are doing so with informal approaches, others are using 
logic, a few are using category theory.  I'm partial to the logic 
approach.  No one seems to have much empirical evidence which approach 
works best.

>Note that I have not used the phrase "category
>theory".  Robert K. cited the work on Specware by
>the Kestrel Institute, which also uses category
>theory.  However, if you look at the Kestrel site,
>they don't mention category theory at all.

true, but they do use category theory

>Suggestion:  Robert (or other people on the IFF
>project) must justify IFF based on its results,
>not on the theories it may use to produce those
>results.  I happen to agree that category theory
>is the right framework to use, but they have to
>learn how to preach to the entire congregation,
>not just to the choir.

I agree.

Adam

>John
>
>