Search Mailing List Archives

Limit search to: Subject & Body Subject Author
Sort by: Reverse Sort
Limit to: All This Week Last Week This Month Last Month
Select Date Range     through    

cost of creating GO

Jean-Luc Delatre jld at
Fri Apr 26 00:33:43 PDT 2002

[foreword to posted list: I am copying to the list this reply to 
William L. Jarrold <billj at> off-list comment about GO 
ontologies because I feel that the subject matter ought to shed some 
light on more "general" problems that ANY, would be, ontology user is 
bound to meet in the (not so) long run. Parts of this reply are somehow 
private and will have no sense or interest to other people, but, 
mostly near the end of the mail, there are general interest developments]

Dear William,

*Some* comments on your last message, 
I have not yet probed deeper into the matters.

 [snip] ...

JLD> > I guess I will have to dig thru the [GO] ontologies and make some
> > consistency checks to figure out what.  From what can be seen on the GO
> > mail archive not everything is that easy.
WJ> I am very interested in such consistency checks.  If you have any specific
> ideas about what sorts of consistency checks there should be, I'd be
> interested in hearing about them.

These checks depends most on what kind of redundancies are available.
I have had no time yet to really investigate the GO ontology structure but 
it seems quite crude, closer to a dictionary/thesaurus than to a "true" ontology.

WJ> When Cycorp had a Special Interest Group for Testing, I was its head.
> (I'm assuming you know what Cycorp is, if you don't, lemme know.  And/or
> see

Oh, yes, I do know Cycorp!
The 500 hundred man years project having run for 18 years and always
due to reach human-like abilities in the next five years.  :o))


WJ> I wrote a paper on my testing work for Cyc and presented it at the AAAI
> Spring Symposium entitled Validation of Intelligence.  I have attached
> this paper and would love any and all comments you have on it.

I will do that, a bit later.

JLD> > Other places are just talking (SUO) or piling up masses of junk
> > (Semantic Web).
WJ> Yes, I've been a member of the SUO since around Feb and randomly sample
> email on that list. 

Well, hopeless, too much of theoretical concerns and little considerations for 
the *real* needs of users (which, BTW, don't seem to be too precisely targeted)
I derided this as the ALGOL68 syndrome:


Since you probably missed most of the "fun" from SUO, I will give you
in this mail plenty of references to interesting SUO postings.

WJ> You mentioned the semantic web.  I am even more interested in hearing
> more about why you think the semantic web is piling up masses of junk.  

The Semantic Web is operating just like Cyc and with 
an even less "solid" approach: 

 "The RDF notation, which was originally designed by R. V. Guha, 
  who was formerly the associate director of Cyc, is a trivial 
  subset of what Cyc was doing in 1984."

Quoted from:

 "I believe the WWW-community's "ontology" stuff to be pretty bad, 
  bordering on junk, and dangerous to the goals of the ontology 
  community in general. But it's there and gaining momentum, 
  because they have TOOLS and CODE and EXAMPLES and not just TALK."

Quoted from:

So, what do you expect?

I expect, myself, that the Semantic Web will meet the same problems
that Cyc met or worse, except SCALED UP TO THE SIZE OF THE WEB!!!

Isn't that an interesting perspective?

WJ> I am in the process of revising the paper I have attached for the purposes
> of a IAAI Workshop this summer called "Ontologies and the Semantic Web".
> Your examples of semantic web junk, in addition to your opinions and
> feedback are likely to be an important part of the paper I am writing.

Sorry, I am not an academic and not craving for such fame.

 [snip] ...

WJ> Well, whether or not there is a misunderstanding here
> versus a typo, your points are well taken. I am going to assume that your
> use of "IA" refers to AI and/or SIA's.

Yes, of course, a typo!

 [snip] ...

JLD> > P.S. What is *your* interest in ontologies?
WJ> I worked on Cyc for 10 years as an Ontological Engineer/Programmer.  Over
> time, I basically came convinced that there was a HUGE need to test the
> system constantly.  New work was interfering with old work to an unknown
> and apparently significant extent.  Towards the end of my stay at Cycorp
> my mostly automatic nightly testing system was finding one new KB related
> bug (i.e. not a bug related to inference engine code) per day.  OE
> methodology seriously needs addressing.  Testing/Validation is the most
> sorely needed missing piece to methodology development; OE's need
> feedback about how there new work breaks old work.  Such feedback will
> hopefully teach them to write more re-usable less brittle OE.  In the
> meantime, they are flying blind.

Ha! it will take MUCH more than "Testing/Validation" to overcome
potentially devastating trouble in ontology uses.

I say "devastating" because the problems encountered will be akin
to software bugs and software maintenance problems and because any 
bit of knowledge in an ontology has far more widespread reach than 
just a line of source code, as badly crippled it might be.

So "Validation" is going to be overwhelmed by the sheer volume and
complexity if consistency is not targeted from the very early
knowledge acquisition phases.

This will suck up the TIME and RESOURCES of countless people and organisations.

This is why in my opinion there should not ever be any 
such thing as an "Ontological Engineer"! (sorry Bill...)

What I mean is, that having to have an "intermediate" level of processing
between the domain expert and the ontology seriously choke down the bandwidth
of knowledge acquisition, and, WORSE, *loose* valuable information:


Though I am bitterly "fighting" Sowa on philosophical and metaphysical
grounds I nearly always agree with him on all practical points.

An important point is the underlying "hidden" assumption for Cyc, SUO,
the Semantic Web and any other of the same vein, that there exist 
ultimately a *perfect* all encompassing ontology (even if allowing
for some "fuzzyness") that allows describing  everything, 
and we have to chase for it however distant it be.

This is absolutely FOOLISH and the root of all "ontology evil"!

Every ontology pertains and belongs to a given *observer*, 
be it a person or a group, and will (rightfully) contain some 
idiosyncratic informations.

The point is not to MERGE all such ontologies, 
which is an endless and silly quest, but to ensure that communication 
between two parties, mostly accustomed to two different ontologies, 
reach an ADEQUATE level of understanding.

The concept "adequate" depending on the purpose of the communication
and therefore, not itself realistically subject to an UNIQUE definition.

What I have been suggesting is an agreement, NOT on the ontologies 
contents for any field of knowledge whatsoever, but on the protocol
used to reconcile meanings of a given concept taken from two 
distinct ontologies.


I elaborated a little more on that (see the "More details & more..." links in
my english version home page

The main points are:

- To ensure proper communication between remote parties their 
  ontologies NEED NOT be identical, not even with respect to
  the revision level of common reference sources!

- Consistency must be taken care of since the very time of data 
  entry and the consistency "proofs" carried along with the
  definitions and assertions. This, to allow receiving parties to 
  check it for themselves WITHOUT any search involved, such as not
  to strain the capabilities of the "proover" on the receive side.

- Common agreement is only to be mandatory for a VERY LIMITED set
  of words upon which all other definitions will be built.
  This agreement need NOT EVEN be on the MEANING of such words
  but only to the fact that those words are required to have
  the same meaning for everyone. Anybody being able to refine
  on the attached connotations for those words, bearing in mind 
  that, if they go astray, others will reject their communications.
  Yet, this allow to have INCOMPLETE definitions (even an empty one!)
  of those words in any ontology .

- Other words, not being required to have a common meaning but being
  definable from the common shared words, will be distinguished by
  their source.
  A 'car' in chinese will be different from a 'car' as we define it but 
  still be made to correspond to the exact concept the author had in mind.

  See an interesting example about "chinese" cars in Sowa's paper:  (Figure 4)

Once the protocol is agreed upon, all agreements about contents can be 
reached thru "cooperative" design by interacting USERS, not "thinkers"!

A final word. 

Natural languages are still beyond the reach of such a scheme as well 
as from ANY other technology reasonably mature as of today.

The point is that "technical" vocabularies and assertions ARE within
reach of the technology sketched above and this is what the current
*realistic* applications are requiring.

So, Bill, you will (possibly) have to reframe your "Ontological Engineer"
skills to a slighly different field. INTELLIGIBILITY of the concepts is
where there will surely be a lot of help needed, because for this, 
NO mechanical trick will do!


-- Jean-Luc Delatre
"Everything should be made as simple as possible but not simpler." 
    - A. Einstein
-------------------------------------------------------------------  -- GSM: +33 6 11 24 06 29

This message is from the GOFriends moderated mailing list.  A list of public
announcements and discussion of the Gene Ontology (GO) project.
Problems with the list?           E-mail: owner-gofriends at
Subscribing   send   "subscribe"   to   gofriends-request at
Unsubscribing send   "unsubscribe"  to  gofriends-request at

More information about the go-friends mailing list