A few replies to comments by John Sowa and Ali Hashemi:
[JS] >>
> But I have been asking Pat the following questions for several years,
> and I would ask anybody with a new proposal:
>
> How does your proposal or suggestion differ from what Cyc has
> been doing since 1984? If you started with Cyc as a base, could
> you proceed faster? Or would something in Cyc hold you back?
> If so, what?
> (01)
[PC] Cyc is (1) mostly proprietary, and a public language is essential (2)
no effort by a single group can demonstrate interoperability among
independent development groups. The notion of having a detailed
well-axiomatized FO to relate domain ontologies is somewhat similar to what
CYC has done with its BaseKB (used by all its domain ontologies), but (3)
CYC has not attempted to use the BaseKB as the sole unifying standard of
meaning - so CYC is not a proper test of this tactic.
The fundamental parts of CYC would be an excellent base for a common
foundation ontology, and if all of CYC were made free and public, it would
doubtless accelerate the development of a widely used common FO, **if and
when** some large diverse community of developers and users organized a
proper collaboration. I have not seen anything in the CYC base KB that
would be contradictory to anything I would want to see in an FO, though I
would want to add some elements that are not contradictory to the CYC BaseKB
- and I expect that other groups would also find some things that they want
in the FO that are not in CYC. I (and others) find some things in CYC that
seem odd or unne3cessary, but none of those I have seen *contradict* what I
would have in an FO - they just say things in a different way. As I have
mentioned, I do not fear redundant alternative representations of the same
intended meanings because I anticipate that any practical application will
*select* from the FO only those parts that are required for the local
application, to minimize the size of the working ontology and improve the
efficiency of the reasoning. In building the COSMO, I have found that a bit
more than half of the elements I consider useful for a basic FO are already
in CYC (or something close in meaning), but there are some things missing.
But I expect that starting with the full CYC would only eliminate a small
fraction of the effort, because most of the work of building a common FO
will be the alignment of or specification of local domain ontologies with
respect to the FO, not in building the FO itself. The technical task of
building an FO satisfactory to a wide diverse community is only the first
part of the project I think is needed; most of the work is in aligning local
ontologies with the FO, building useful applications that use those local
ontologies, and demonstrating that alignment with the FO does indeed support
the accurate communication of information among the independently developed
applications.
A related question that John has raised in the past is, if the CYC
approach is technically adequate to support effective AI applications, why
do we not see such applications? If we assume that even the proprietary
applications that CYC has built (not available for public inspection) are
also not particularly impressive, there are possible reasons other than that
the approach itself is flawed. The most likely, I believe, is just that the
effort of **using** an ontology to good effect (not just as a minor
improvement on database technology) probably requires substantially more
effort than building the ontology itself. I not only haven't seen anything
from CYC that is public and impressive, I haven't seen anything from anyone
else that is public and impressive that uses ontology to do things not
easily done by other techniques (and there are *many* ontologies around). I
do **feel** (from small thought-experiments and trivial tests with
logic-based ontologies) that the kind of detailed hand-crafted ontology that
CYC builds is essential to approach human-level reasoning; and I suspect,
based on the history of the past twenty years, that nothing much short of
human-level reasoning will be impressive enough to elicit much more than a
ho-hum from IT managers. (John's "challenge" suggestion of talking to a
six-year old in fluent English would be one such impressive example). There
are effective non-ontological ways to mimic human expertise within a small
domain; the big problem is to support intelligence in multiple domains and
allow accurate transfer of information among them.
As for whether CYC has "failed", they are currently getting most of
their revenue from commercial projects. (02)
> PC> I do think that the set of concepts represented by the **defining
> > vocabulary** (2148 words of English) used in Longman's is analogous
> > to the concept representations of an FO; because of ambiguity, Guo
> > estimated that the number of different senses used in the
> definitions was close to 4000.
>
> [JS] You have made that claim many times, and the only thing you can point
> to is Longman's dictionary. But the definitions in that dictionary
> are hopelessly vague. It would be impossible to do any kind of
> serious reasoning with those definitions.
>
[PC] Many of the actual definitions in Longman are relatively simple, but
I have tried some experiments and have concluded that one can make those
definitions as detailed and specific as one wants, ***and still use only the
words in the Longman defining vocabulary (LDV) (or words defined
recursively, grounded in the defining vocabulary)***. I have created
definitions for 600+ words not in the Longman DV, and those definitions with
their trace back to the LDV can be found at: (03)
http://www.micra.com/COSMO/DefiningVocabulary/SupplementalVocabularyDefiniti
ons.xls (04)
I have in the past suggested that if anyone doubts the adequacy of the
Longman vocabulary to properly define (in English) any terms of interest,
they should prepare what they consider to be a proper definition of some
term, and I will try to demonstrate how the words of that definition can
themselves be grounded recursively in the LDV. Only one person has taken up
that suggestion, and I did send what I consider to be adequate definitions,
with no follow-up. The offer is still open. The ultimate goal would be to
have English definitions based on some defining vocabulary that could be
*automatically* converted into the proper logical specification of meaning.
I have not even begun to create such an English interpreter, there is a lot
of work to do before that. As for logical definitions based on an FO: (05)
> If you want to make a convincing argument for those primitives,
> I suggest the following exercise:
>
> 1. Select some small, but reasonable subset of words from the list
> (say about one or two dozen), which could be used to define
> another dozen words that are not in that list.
>
> 2. Give formal axioms for the defining words in some version
> of logic, preferably Common Logic, but you can use any other
> formalism you prefer.
>
> 3. Then show how that formal specification could be used to
> define the dozen words not in that list.
>
> Before you urge other people to do this kind of thing, you should
> demonstrate how to do it yourself.
>
[PC] If in fact anyone would be **convinced** (which I seriously doubt) by a
small demo of that type I will be happy to spend the effort to do that, too.
But I suspect that to do most such definitions would require that most of
the FO would already have to be specified in FOL. At present the COSMO is
only an OWL 1.1 version, without any rules that cannot be expressed by the
built-in OWL logical functions. Still, I am game to try. But to avoid a
time-costly and pointless effort that will fall on deaf ears, I would need
some kind of commitment from some skeptic to **admit** that this approach is
seriously worth investigating, if such a small demo succeeds. If anyone has
a suggestion for an ontology element whose logical specification in terms of
the logical primitives of an FO they would find impressive, please let me
know which one. From past experience, I feel it likely that anything I
choose would be dismissed as not convincing. (06)
>From Ali:
> AH> As it stands we have many people who are working to develop this
> > interlingua; we are in effect, defacto developing exactly the set
> > of primitives you speak of, except in a not very coordinated manner
> > and without an overarching framework. While this lack of cohesion
> > introduces some problems, it also means work can progress without
> > waiting for consensus.
>
Which project are you referring to? Without a pointer to publicly
observable details I can't respond meaningfully. (07)
And again:
> AH> I will only note that a functioning family of interlingua
> ontologies
> > - a collection, perhaps situated within a repository would suffice
> > for almost all purposes of intercommunication. It makes the question
> > of producing a single foundation ontology somewhat superfluous, with
> > quickly diminishing returns on value.
>
[PC] I don't know what you consider "an interlingua ontology". I am all
for repositories, and there may be many domain ontologies that can be reused
outside the shop of the originators, that can be useful for related local
applications. But unless the repository itself includes detailed mappings
between ontologies that support **accurate interoperability**, I cannot
visualize how such a repository would make a common FO "superfluous". More
detail?? In what way would a repository per se enable interoperability? (08)
I recall at the 2009 ICBO conference in July that there were some efforts
to begin the process of actually mapping the domain biomedical ontologies to
each other, or to a common higher-level ontology (beyond linking to BFO).
But the efforts at that time were very preliminary, and I did not see any
serious effort at the kind of coordination that would be required to
actually achieve accurate interoperability. I hope there has been some
progress since then, and if you can relate the details of such an effort, or
of some other related effort, I expect that it would be quite interesting
for this discussion list. (09)
Pat (010)
Patrick Cassidy
MICRA, Inc.
908-561-3416
cell: 908-565-4053
cassidy@xxxxxxxxx (011)
_________________________________________________________________
Message Archives: http://ontolog.cim3.net/forum/ontolog-forum/
Config Subscr: 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 join: http://ontolog.cim3.net/cgi-bin/wiki.pl?WikiHomePage#nid1J
To Post: mailto:ontolog-forum@xxxxxxxxxxxxxxxx (012)
|