Re: [ontolog-forum] Axiomatic ontology

[Pat Hayes <phayes@xxxxxxx>]

At 8:38 PM +0200 1/31/08, Avril Styrman wrote:

> Pat,
>
> I had similar thoughts than Rob here:
>
> Rob Freeman:
> > I don't see why it should not prove possible to reason with
> classes so
> > defined. I believe indeed that natural language can be thought of
> as a
> > formal system over such classes (a formal system as distinct from
> a
> > grammar.)
>
> Comments below.
>
> Lainaus Pat Hayes <phayes@xxxxxxx>
> > >1:
> > >Human is a mammal
> > >
> > >2:
> > ><class> Mammal </class>
> > ><class subClassOf:Mammal> Human </class>
> > >
> > >A NL parser can extract the same information (human is
> > >a mammal) from NL that can also be given in a formal
> ontology.
> >
> > NL parsers cannot extract information. One needs
> > also NL semantics. Which of course amounts to an
> > ontology. This whole topic of NL understanding
> > belongs in AI. One needs a basic acquaintance
> > with the subject before forming a useful opinion.
>
> Take an RDF Schema. The class and property structure
> is usually in one file, and the instances in another
> file.

Well, no, in fact. Not that it matters which file something is in.

> You say that the instance-file is not a
> part
> of the ontology, but it is only a knowledge repository.

No, I didn't (and don't) say that. (Are we on the same planet??)

> Ok, I can go with that.
>
> But, the is-a relationships between the classes surely
> must be parts of the genuine ontology, isn't this what
> you say?

Yes, I'm willing to go with that.

> It does not require millions of dollars
> to implement
> a NL parser that searches "X is Y" phrases from the
> web.

Indeed, but (1) many sentences of this form are not about class-subclass relationships , and (2) many subclass relationships are expressed in other ways, and (3) not all assertions which can be thought of as expressing such a relationship are in fact best understood that way.

> All the parsed results can be compared,
> and
> probablilities given to each individual X and Y in
> "X is Y". And this is exactly the same as making
> a 'genuine' ontology with the subclass relations.

No, it is absolutely nothing like making a genuine ontology.

> It is only a domain-independent way to do
> it,
> when hard-coding is not, unless you hard-code the
> whole world in that box.
>
> If you do not consider that even the subclass
> relations are parts of a genuine ontology, then

Forget the rest of this, as I *do* think that subclass relations are part of an ontology.

> ...
> > Your comparison above is childishly simple, and
> > does not properly reflect the real problem, which
> > has remained intractable for several decades with
> > many very clever, highly motivated, people and
> > with millions of dollars in funding, trying their
> > very best to solve it.
>
> The is-a relationship is a part of the solution.

No, the problem is to find the ontological content in the free NL text. Such text contains all sorts of information, some of which can be possibly expressed as subclass relations, and a much smaller fraction of which can be usefully so expressed. But there are no simple heuristics for reliably recognizing such information based on the form of the NL sentences.

> If you don't agree, then I don't know
> what is your
> position.
>
> A complete solution would of course be a machine
> that gives an answer to any question that has
> an answer. Are you asking now how the solution
> can be programmed? I can give some very general
> guidelines, such as Zadeh's PNL.

I'm sorry, but it is clear that you know almost nothing about the actual NLP field. There is no point in continuing this discussion. Read some textbooks on the state of the art on NL comprehension systems.

> > The Web is NOT an ontology. If you
> think I am
> > wrong, don't argue with me: go ahead and
> > implement a tiny fraction of the needed
> > functionality, and you will be richer than
> > Croesus.
> >
> > >
> > >E.g. Lotfi Zadeh's Precisiated Natural Language PNL
> > >is all about using the internet as the source of
> > >knowledge, and turning it into a formalized
> form.
> >
> > As are many other projects, including IBM's
> > multimillion-dollar WebFountain project. None of
> > them will ever come remotely close to being able
> > to extract coherent structured ontological
> > content from free text. The best one can do is to
> > 'scrape' some approximate shallow, very
> > unreliable, propositions and to identify things
> > like proper names of people and cities. This is
> > of course extremely useful, but it does not come
> > anywhere near being able to treat the Web as an
> > ontology.
>
> PNL is a general scheme. It does not solve all the
> problems of AI, but it is a good suggestion of
> the general guidelines.

The general guidelines for solving all the problems of AI?? I'm afraid I am simply laughing at this point.

> > >This way, the naturally evolving
> internet contains
> > >similar sort of information than that what is
> > >inserted into Cyc by hand.
> >
> > No, it does not. If only because the Web does not
> > comprise simple declarative sentences of the kind
> > found in reading primers.
>
> One of Zadeh's ideas was to convert NL into PNL.

And the best of luck to you and to Lotfi in actually managing to do this.

> I suggest you read the article:
>
> AI Magazine  archive
> Volume 25 ,  Issue 3  (Fall 2004) table of contents
> Pages: 74 - 91 
> Year of Publication: 2004
> ISSN:0738-4602

Ive read it, of course. But AI Magazine is to real research as Scientific American is to a publication in a professional physics journal. When Lotfi gets some actual results and publishes them, I might take some serious notice.

Pat Hayes

> Avril

-- 

---------------------------------------------------------------------
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

_________________________________________________________________
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    (01)