ontolog-forum
[Top] [All Lists]

Re: [ontolog-forum] Integrating Semantic Systems

To: ontolog-forum@xxxxxxxxxxxxxxxx
From: "John F. Sowa" <sowa@xxxxxxxxxxx>
Date: Sat, 05 Jun 2010 13:08:50 -0400
Message-id: <4C0A84A2.40102@xxxxxxxxxxx>
John, Pat, and Mills,    (01)

Thanks for the comments.    (02)

JB> The web site for Controlled Natural Languages has changed to:
 > http://sites.google.com/site/controllednaturallanguage/    (03)

Thanks for the reference.  That is a major problem with a pointer
based system:  pointers are frequently broken, and it's unclear
what the pointers point to.    (04)

One of the articles cited there surveys a wide range of controlled
NLs and evaluates various problems and successes:    (05)

    http://utilika.org/pubs/etc/ambigcl/clweb.html
    Can Controlled Languages Scale to the Web?    (06)

But I would note that every problem he notes about CNLs also occurs
with any notation of any kind.  Many people say that Linked Open Data
(LOD) is "The Next Big Thing" -- but LOD has no semantics, and it
doesn't solve the pointer problems.    (07)

My claim is that CNLs provide a readable way to represent any
semantics of any kind.  If you have a system with well defined
semantics, controlled NLs can make it readable.  But if you have
no semantics, they have nothing to express.    (08)

Unfortunately, LOD is a triumph of syntax over semantics: everything
is accessible, but the only semantics is in the mind of the reader.
It is an implementation of the Library of Babel:    (09)

http://jubal.westnet.com/hyperdiscordia/library_of_babel.html    (010)

PC> In your slide 28, the SQL query does not appear equivalent to
 > the CLIF assertion:
 > ---------  Slide says: ----------
 > SQL query:
 > SELECT FIRST.ID, SECOND.ID
 > FROM OBJECTS FIRST, OBJECTS SECOND, SUPPORTS
 > WHERE FIRST.TYPE = "Cat"
 > AND SECOND.TYPE = "Mat"
 > AND SUPPORTS.SUPPORTER = SECOND.ID
 > AND SUPPORTS.SUPPORTEE = FIRST.ID
 >
 > Common Logic Interchange Format (ISO 24707):
 > (exists ((x Cat) (y Mat)) (On x y))    (011)

Yes.  SQL requires the user to know the choice of tables that the
DB administrator happened to choose.  When I wrote that query, I
was thinking of tables similar to those in slide 87.  (But in
an earlier version, the columns of the Objects table were named
ID, TYPE, COLOR.)    (012)

PC> The non-equivalence is due to the possibility that there will
 > be subtypes of "Cat" (e.g. SiameseCat)  and "Mat" (e.g. StrawMat)
 > in an ontology queried by FOL, whereas the SQL suggests that an
 > Object has only one unique Type.    (013)

The need for a type system in RDBMS was well known back in the
1970s.  Ted Codd and Chris Date (among many others) had detailed
proposals to address that issue.  See slides 16 to 18 for some
discussion about the conceptual schema.    (014)

PC> If the English statement is to be interpreted as meaning that
 > the immediate declared type of the cat must be "Cat" and not
 > any subtype, then the issue of multiple inferred types may not
 > exist here.    (015)

If you have a type hierarchy, then the type Cat is true of anything
declared to be an instance of Cat or any subtype of Cat.  Since
the users cannot know which subtypes have been declared, they
should be allowed to mention any type they find convenient
(including supertypes such as Cat, Animal, or Something).    (016)

People were very well aware of these problems in the 1970s.
Just look at slides 17 to 21 in the following talk:    (017)

    http://www.jfsowa.com/talks/cnl4ss.pdf
    Controlled Natural Languages    (018)

Those slides discuss the TQA system.  Back in 1978, TQA was
tested in a year-long study with land-use planners who had
no skills in programming or DB design.  And they loved it.
But IBM killed the project because the expense of defining
semantics was too great for a profitable commercial system.    (019)

Today, after 32 more years of R & D, LOD has *less* semantics
than the RDBMS of the 1970s, and it has much, much, much less
semantics than TQA from 1978.    (020)

For LOD, people are using RDFa, which is supposed to be RDF
with "attributes".  But the attributes are undefined character
strings, and the only similarity to RDF is the substring 'RDF'.
The reason for that substring is to disguise the painful fact
that the Semantic Web was a failure:  the only thing that
survives is an extension to HTML for adding undefined tags.    (021)

PC> There are a couple of  typos in slide 33:    (022)

I copied that slide from a talk by Pat Hayes.  I'll revise it.    (023)

MD> Last fall you called attention to Pat Hayes' keynote presentation
 > at ISWC 2009 entitled "Blogic", in which he outlined suggestions for
 > addressing shortcomings of RDF (and OWL) with some approaches (e.g.
 > surfaces) that would align semantic web standards more closely with
 > CL. I'm not sure of the status of development and discussion relating
 > to RDF 2.0, but suggest that a synopsis of  latest developments would
 > be appropriate for your tutorial.    (024)

Good question.  My impression is that Pat's talk made a splash at the
conference, and the waves have long since dissipated.  If anybody
has more info, please let me know.    (025)

John    (026)

_________________________________________________________________
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    (027)

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