Saturday, April 22, 2006
Community centric
service methodology à [home]
Regarding
http://www-db.research.bell-labs.com/user/pfps/publications/architecture.pdf
and related issues.
The text below is
slightly edited and re-formatted for easy of reading (psp)
Rolf and Avril,
I wouldn't say that the Semantic Web is totally bad,
just as I wouldn't say that the original IBM PC was totally bad. In fact, the PC had three very, very good
qualities: the steel case (instead of
the plastic case of the aptly named Trash-80), the full keyboard (instead of
the upper-case-only keyboards of the TRS and Apple PCs), and a decent, if not
very exciting, monitor that displayed 80-characters across (instead of the 40
chars of the TV-based displays).
But the IBM Instruments Division in 1980 had produced
an engineering computer based on the Motorola 68000 with an operating system
and compilers written by IBM. It was built into a bigger and uglier box with
seven expansion slots for adding instruments and process controllers of various
kinds.
In 1981, it would have been easy to put a scaled-down version of the 68000 system into exactly the same box as the PC. IBM could have delivered it with an OS and compilers from IBM instead of QDOS (Quick and Dirty OS -- which Bill Gates bought for $50K and sold to IBM for one million as MS-DOS 1.0).
That's my feeling about the Semantic Web. It has some good features (such as Unicode, URIs, and the use of XML for marking up documents, not for implementing languages). But it is so much worse than it could have been if the developers had taken account of the current state of the art (in commercial use, even, not to mention research).
RS
> Before you get too sick, have a look at:
> http://www-db.research.bell-labs.com/user/pfps/publications/architecture.pdf
>
> A kind of fix :-) .
Thanks for the pointer. Peter makes many of the same points I made:
· RDF is a terrible base for supporting logic,
· OWL demonstrates how bad RDF is, and
· there was much better technology available instead of the bunch of crap that was dumped on us by the W3C.
The main difference is that Peter stated his points in a more diplomatic way than I did, but I agree with his conclusion:
PFPS
> It would also be possible to generalize the revised
> Semantic Web architecture in several ways. One could lift
> the requirement that systems handle lower levels of the
> stack, turning the Semantic Web stack into a collection of
> languages with a common semantic framework. One could also
> loosen the requirement of a common semantic framework into
> simply some sort of semantic compatibility.
My recommendation is to adopt the draft ISO standard for Common Logic as the foundation for the Semantic Web. There are already mappings of RDF and OWL into CL, and there is also an XML notation, called XCL, defined in Annex C of the standard. As for the use of XML, I would make that optional and provide an XML tag with the option LANG=xxx in order to support better language formats.
AS
> ... It is easier for a normal person to create an ontology
> with a frame-based ontology editor than by using SQL tables.
> Sure, the idea of frames existed long before SemWeb, but after
> working with a large and messy SQL database for two long years,
> I would welcome a frame-based SQL editor. That would surely help
> to build and to use a database, since SQL databases are (imho)
> very much easier to apply than RDF databases.
I certainly agree with the two main points: SQL does not have good tools for supporting a type hierarchy, but SQL databases are far easier to use than RDF databases.
But you don't need frames or RDF for a good type hierarchy. Aristotle developed that over 2300 years ago, and Porphyry developed tree diagrams for representing type hierarchies over 1700 years ago. Ted Codd and Chris Date were arguing for years that SQL should support types.
I blame IBM and Oracle for the weaknesses of SQL. First, IBM is to blame for not adding types [1], which were being discussed by many people (including me) within IBM during the 1970s. My first published paper on conceptual graphs in 1976 proposed CGs as a schema language for RDBs:
http://www.jfsowa.com/pubs/cg1976.htm
Second, I blame Oracle for implementing the first version of SQL that had been published in the IBM Journal of R & D instead of the much better QUEL language that Stonebraker had developed for Ingres.
AS
> SemWeb languages have at least two good features: they
> are frame-based and they are easily shareable. I'd be better
> off with SQL that would accommodate these two features. Then
> again the errors of RDF appear while one is trying to use
> an RDFS ontology by working with e.g. JENA. It is slightly
> amusing, but mostly painful.
I agree. There's an old saying,
Those who don't know history are doomed to re-implement it.
John
[1] Footnote to clarify “type”
I do not recommend the word
"class" because there is
no commonly accepted
definition, and any use of that
term is an open invitation
to confusion.
I recommend the following
distinction:
-- Sets: X is a subset of
Y in the context (or model) C
if and only in the context (or model) C, every instance
of X is also an instance of Y.
-- Types: X is a subtype
of Y if and only if for every
context (or model), every instance of X is also an
instance of Y.
In short, sets are context
dependent. The set of all human
beings changes with every
birth and death, but the type Human
remains the same. If somebody uses the word "class",
you
don't have a clue about what
they mean.
If you must use OWL notation,
I recommend that you avoid
the terminology used in the
OWL documentation and adopt
the words "set" or
"type" to indicate what you intend.
Of course, you should also
include a footnote to indicate
that the OWL terminology is deprecated.