[Top] [All Lists]

Re: [ontolog-forum] Ontology and methodology

To: "Cassidy, Patrick J." <pcassidy@xxxxxxxxx>
Cc: "[ontolog-forum] " <ontolog-forum@xxxxxxxxxxxxxxxx>
From: Pat Hayes <phayes@xxxxxxx>
Date: Tue, 20 Mar 2007 14:32:52 -0500
Message-id: <p06230902c225dc7745c4@[]>
>Pat Hayes raised several issues I think are worth further discussion,
>>  But what could possibly be in such an ontology? Consider a Cyc-like
>>  effort, aimed at 'common sense'. Now suppose your domain of interest
>>  is navigating a robot submarine through deep ocean water. You will
>>  need many purely spatial concepts, but I will lay very good odds that
>>  no more than one or two of those in the Cyc-style Kbase will be the
>>  slightest use. You will have to re-think what it means to 'rotate',
>>  for example, or what 'outside' means. Most of the everyday spatial
>>  concepts will simply be useless when swimming in a deep ocean. (This
>>  example actually came up when I was working on the Cyc project, and
>>  we checked.) Or suppose your domain is describing how viruses get
>>  inside human cells by tricking their membrane chemistry: will the
>>  notion of 'boundary' you are using really be similar to the notion in
>>  the foundation? If the latter is based on mathematical ideas it won't
>>  be, because the membrane, at this scale, has an elaborate 3-d
>>  structure. How about 'solid'? Sorry, at this scale the membrane is
>>  solid-ish perpendicular to its immediate tangent plane, but more like
>>  a liquid in that plane, so its vesicles can float freely around the
>>  membrane. Its an odd kind of thing that doesn't exist at the everyday
>>  common-sense scale at all. Will that be in your foundation ontology?
>>  Everywhere you look at real domains, one finds new kinds of 'thing'
>>  that you wouldn't have thought of unless you were working in that
>>  domain. (The Horatio principle: there are more things in heaven and
>>  earth than are dreamt of in your ontology.) This is basically why
>>  these universal ontologies never succeed: nobody has a rich enough
>>  imagination to think of all the things that might be needed, which is
>>  another way of saying all the counterexamples to any philosophically
>>  motivated distinction.
>Yes, there are tens of millions of categories of things and attributes
>and relations that people will want to talk about in various contexts,
>and most of those will not yet be in Cyc, but the question is whether
>there is a much smaller set of basic concepts whose meanings are quite
>well defined, and which serve to provide the basis for combinatorial
>description of all the more complex concepts.    (01)

If that really is a question, I will argue that 
the answer is NO. This is a very old dream, going 
back at least to the French Encyclopediaists and 
maybe further, of there being a single basic 
stock of Ideas from which all others can be 
constructed. Every attempt to compose such a list 
has failed, and there is no evidence at all, as 
far as I can see, that this list of 'basic' 
concepts exists, or that we could possible find 
it even if it does. (If it existed, why would the 
methodology you propose be more able to find it 
than the one used by CYCorp?) There isn't even a 
useful coherent notion of what it means to 
"combine" concepts into larger ones: all the 
ideas along these lines emerging from psychology 
are flawed or way too simple for real use. (One 
exception might be the idea that all 'natural' 
concepts are formed by analogical/metaphorical 
mappings from a basic stock of proprioceptive 
body-image-based concepts. But if this idea is 
true, we have to trash almost all extant work on 
ontology and start again in any case.)    (02)

I'm not arguing that there aren't some generally 
useful 'low-level' ontologies. Time and 
geographical space are obvious topics that we 
ought to be able to 'get right' in a reasonably 
useful way. Time in particular seems lot easier 
that space, and has had a lot more work done on 
it, and there are now "standard" temporal 
ontologies such as OWL-Time. But let me tell you 
a true story. I recently started collaborating on 
a project to "ontologize" what artists say about 
their artwork. Ignoring details, we collected 
transcripts and tried to formalize obvious 
low-hanging conceptual fruit. One of the first 
ideas we found was 'golden hour' , a 
photographers term for the time before sunset 
when the light has a certain quality. This seems 
like a pretty obvious time-interval, right? 
Wrong. There is nothing, repeat nothing, in any 
of the extant temporal ontologies that can be 
used to describe the golden hour. So, back off, 
and try 'late afternoon'. Again, nothing. How 
about 'dawn', referring to the time of the 
sunrise? Nothing. Lots about the time zones and 
calendars and leap years, nothing about dawn or 
sunset.    (03)

This kind of thing is BOUND to happen. The only 
way it would not would be if the people writing 
the base ontology could think ahead of every 
possible application and every possible domain, 
to see what will be needed. Nobody can do this.    (04)

>   If we agreed on that
>basic set of concepts (and called it the 'foundation ontology' or any
>other term) that would allow all of the wonderful diversity of
>ontology-building that we all agree can lead to great things, but it
>will also provide us all with a good mechanism to relate these
>ontologies to each other, by specifying the meanings of those terms and
>concepts (to whatever degree of detail is needed), using the same set
>of basic terms from the common foundation ontology.   Any required
>fundamental concepts found to be missing from the foundation ontology
>will be added when they are discovered.    (05)

All concepts are 'basic'. It is rare, and only 
occurs in artificial situations, for one concept 
to be *definable* in terms of others. There is no 
foundation: or, perhaps, the entire building is 
foundation.    (06)

>There is already an existence proof of just that mechanism, in the
>"controlled defining vocabularies" used by some dictionaries to define
>their terms; with about 2000 defining terms, Longman's can provide
>meaningful definitions of over 100,000 dictionary terms.    (07)

With that many WORDS. But most concepts have no 
corresponding word. If ontologies only needed the 
concepts expressed in English words, ontology 
engineering would be a branch of literature. And 
the kind of definitions that one finds in a 
dictionary are nowhere near as precise or 
unambiguous as needed for ontology work.    (08)

>  Since those
>words are labels for concepts, it makes sense to imagine that a
>comparably small "conceptual defining vocabulary" could be used to
>logically specify the intended meanings of the hundreds of thousands of
>complex concepts that people will want to work with in a computer.  But
>we don't yet know the necessary size of the corresponding conceptual
>defining vocabulary (doubtless larger, since many of the defining words
>had multiple senses), and I am not aware that Cyc or any other
>ontology-builder has intentionally set out to create that conceptual
>defining vocabulary for that specific purpose.  I saw one paper by
>Porter who looked into a similar tactic on a small scale, but does not
>seem to have followed up.  Perhaps someone here is aware of other
>efforts specifically in that direction?
>A critical point here is that it is imprudent to just assume that
>something can't be done when the benefits of doing it can be great.  It
>should be given a  proper chance to succeed.    (09)

20 million dollars is a pretty large sum to bet 
on something this risky. If I were a venture 
capitalist I wouldnt give y'all a dime for this 
project.    (010)

>  The mostly volunteer
>processes used up to now have not worked, but there is still a
>plausible method not yet tried.
>I imagine that the minimum number of basic concepts needed will be at
>least 5000 (types + relations), possibly 10,000?    (011)

What can *possibly* be the basis for an estimate like this?    (012)

>  I am starting to
>explore the question by creating a merger of some of the basic elements
>of Cyc and SUMO (alas, first in OWL, for a couple of practical reasons,
>but necessarily either into FOL or supplemented with rules in some
>form, soon).  The starting point will be about 3500 classes and 300
>relations, and the first question will be how much that will have to
>expand to be able to specify the meanings of the 2000 words used as the
>defining vocabulary in Longmans.    (013)

Well, try defining "handle", or "cover" (a 
famously complicated CYC effort.) A dictionary 
definition is no use: you need something that can 
support a reasonable set of actual inferences.    (014)

>  Beyond that, there are some
>additional standard vocabularies, and a test could inquire how well
>those meanings can be specified.  That should provide some useful data;
>if an asymptote on new basic terms is approached, this will suggest the
>practical feasibility of trying to get agreement on using some basic
>set of defining concepts.  This does not in any way restrict people
>from doing anything they want any way they want to - it merely provides
>a means for them to relate what they are doing to what others are
>doing.  The goal is similar to what I interpret as your goal with
>wide-open module usage on the semantic web, but by having one firm
>grounding of agreed meaning in a coherent and logically consistent unit    (015)

But this is (I think) impossible, or at any rate 
extremely hard, to construct; and more to the 
point, provides no significant utility. It is 
just one more module among many. What decides the 
utility is the uses that people put it to. Since 
that will be decided more reliably and more 
quickly by publishing each piece as soon as it is 
written, why not do that? If you explore the 
current SWeb using Swoogle, you will find that 
quite a lot of useful stuff has already been done.    (016)

>(which can incorporate any modules that find a constituency), I think
>it will make that goal a lot easier to attain.  Logically compatible
>but different-looking representations of the kind you found
>translations for in the IKRIS project can be incorporated into such a
>Those who worked on the Cyc project probably will have some useful
>experience to share, whether or not the Cyc baseKB was intended to be
>viewed as a "defining vocabulary".  In any case I hope that members of
>this community will find time to look through the nascent "conceptual
>defining vocabulary" (I will suggest it as the starting COSMO ontology
>to the COSMO working group) to point out problems or missing elements.
>It would be a lot easier with some funding, but first I need to post
>the starting ontology and provide some data on how well it serves as a
>base.  Being OWL, at this point it is little more than a taxonomy.  I
>hope to have the first OWL version online at the ONTACWG site by the
>time of the Ontology Summit.    (017)

Well, Im sure it will be interesting. But I think 
your main ambitious goal is doomed to failure 
before it even starts.    (018)

>>  The way to stop the debates is not to legislate one side as the
>>  winner (that just changes debate into open warfare) but to allow
>>  everyone to write their ontologies in the way they find congenial
>>  (informed by a basic knowledge of good engineering practice, of
>>  course), and to achieve inter-operation by re-use and translation.
>Right.  I just think that a module the size of the conceptual defining
>vocabulary - which could have separable submodules - will ultimately
>prove more amenable to rapidly approaching the kind of reuse that will
>work - modules that are logically consistent with each other; and if
>they aren't, a way to show that they aren't, so proper precautions can
>be taken.    (019)

I think you are missing the important innovation 
that the SWeb provides, which is concept re-use. 
The proposal isn't to set a bunch of people 
writing ontologies in isolation and then all put 
them on the Web. The idea is that the first thing 
you should do, when setting out to write an 
ontology on a topic, is search the Web for 
existing ontologies on related (or perhaps even 
that very) topic, and re-use the concepts in them 
when possible, perhaps adding to their meaning by 
adding axioms. The result is not a huge 
collection of (likely inconsistent) separate 
ontologies, but a web of related ontologies, 
sharing concepts explicitly and with 'import' 
links (which would be better termed 'assenting' 
links: when I import your axioms, I am 
re-asserting them, agreeing with you explicitly 
in a kind of chorus of assertion.) The result 
really will be a semantic *Web*, not just a bunch 
of ontologies which happen to be in the Internet.    (020)

I think maybe one way for us to agree 
pragmatically is to view the SWeb as a social 
mechanism to enable a distributed effort along 
the lines you want to pursue. You will have to 
use the internet in any case to get the project 
done, right? Then your foundation is a set of 
SWeb ontologies which are warranted to be 
mutually consistent and provide a very widely 
useful (even if not completely universal) 
collection of re-usable concepts. That would be 
worth trying. First though we would need to agree 
on a logical notation to use and set up some 
agreed infrastructure.    (021)

Pat    (022)

>I don't expect that a distributed-base volunteer project will reach a
>usable set of modules any where near as fast as a funded project would.
>Of course, if there is no significant community of ontologists pushing
>for such a project, it will probably never get funded.
>Patrick Cassidy
>260 Industrial Way West
>Eatontown NJ 07724
>Eatontown: 732-578-6340
>Cell: 908-565-4053
>Message Archives: http://ontolog.cim3.net/forum/ontolog-forum/ 
>Subscribe/Config: http://ontolog.cim3.net/mailman/listinfo/ontolog-forum/ 
>Unsubscribe: mailto:ontolog-forum-leave@xxxxxxxxxxxxxxxx
>Shared Files: http://ontolog.cim3.net/file/
>Community Wiki: http://ontolog.cim3.net/wiki/
>To Post: mailto:ontolog-forum@xxxxxxxxxxxxxxxx
>    (023)

IHMC            (850)434 8903 or (650)494 3973   home
40 South Alcaniz St.    (850)202 4416   office
Pensacola                       (850)202 4440   fax
FL 32502                        (850)291 0667    cell
phayesAT-SIGNihmc.us       http://www.ihmc.us/users/phayes    (024)

Message Archives: http://ontolog.cim3.net/forum/ontolog-forum/  
Subscribe/Config: http://ontolog.cim3.net/mailman/listinfo/ontolog-forum/  
Unsubscribe: mailto:ontolog-forum-leave@xxxxxxxxxxxxxxxx
Shared Files: http://ontolog.cim3.net/file/
Community Wiki: http://ontolog.cim3.net/wiki/ 
To Post: mailto:ontolog-forum@xxxxxxxxxxxxxxxx    (025)

<Prev in Thread] Current Thread [Next in Thread>