David and Anatoly, (01)
Since I grew up as a mathematician, I strongly support formal
semantics. But I also worked at IBM on both sides of the R & D
fence, and I realize that *technology transfer* from R to D has
always been a huge problem. Both sides make nasty comments about
the "clueless" people on the other side. Unfortunately, their
criticisms of each other are usually correct. (02)
DP
> I am suggesting that the quickest way for
> semantics/ontology to have the broadest impact is to build for the Web
> first. My point is about things taking decades, which is why I mentioned
> the Semantic Web activity as having energized far more people in the
> past 10 years than anything else. Organizations also get a lot for free
> by starting with a Web-based approach (e.g. universal identifiers with a
> well-understood approach they already understand). (03)
I agree that URIs are useful. But I believe that the Semantic Web would
have been integrated with mainstream IT from Day 1 *if* they had started
with UML diagrams as the human interface and supported *both* RDBs and
web-based data. See the note below from another thread. (04)
DP
> Do a small experiment - Walk into IT depts in companies in agri, bio,
> pharma, defense, mfg, and oil and gas and ask people if they've heard of
> 'ontology' or 'semantics' - most will say no. Of those that answer yes,
> ask them if they have any interest in that subject in their company -
> again, most will say no... (05)
Precisely! But they certainly know UML, which is sufficient to specify
the semantics of the overwhelming number of OWL ontologies published
on the WWW. UML is the *sweet spot* for specifying the semantics of
*both* mainstream IT *and* the data on the WWW. (06)
DP
> Alternative summary: 1) IT is moving more and more towards a set of
> competing platforms. Apple, for example, certainly completely disagrees
> with your assessment. 2) Of the available set of candidate platforms,
> the Internet/Web will be the most long-lived and have the broadest
> impact so choosing that gets the biggest bang for the proverbial buck. (07)
I agree. That is why I recommend UML + controlled natural languages
as the human interface for all the above systems and with formal logic
to define the semantics of all the notations. See the note below. (08)
AL
> 1. Formal semantics. This is necessary but not sufficient. This
> is machine code, only the most brave people in the world create
> on this level, Titans and Heroes. (09)
But I would state the formal semantics in an abstract syntax that
is totally independent of any concrete notation, linear or graphic.
Anybody should be able to use a formally defined language in whatever
notation they prefer. The computer could automatically translate
any notation to any other notation that anybody might prefer. (010)
AL
> Long live Fortran and UML! But, please, live not with me.
> There exist other paradigm languages. (011)
I agree that UML wasn't originally specified precisely, but it can
be mapped to whatever abstract syntax is used for the formal semantics.
For ontology, the two most important UML diagrams are type hierarchies
and E-R diagrams (AKA signatures). Type hierarchies are a variant of
the Tree of Porphyry, which has been used to teach Aristotle's logic
and categories for over 1700 years. E-R diagrams (signatures) are
just as formal. Programmers happily use with both of them. (012)
AL
> 2. Low level language to integrate all (like Assembler: it has low
> expressivity but multiple paradigm languages compiled to this language). (013)
Actually, assembler is the *most* expressive language on any computer.
The reason why it's unreadable is that it maps each high-level statement
to a large number of low-level operations. In fact, I would say that
predicate calculus is the assembler language for ontology. (014)
AL
> 3. Macro assembler or/and subroutines. I want not program in Turing Machine
> language, even in Logic Turing Machine. I want go high. (015)
I agree. But there is no single language that is ideal for all kinds
of applications. So I would support any notation that anybody prefers,
as long as it has a precisely defined mapping to the formal semantics. (016)
As one option, I recommend controlled natural languages, which can be
used just as precisely as any other notation for logic. In particular,
UML has a version of FOL with a notation called OCL, which is equally
repugnant to both programmers and logicians. I would replace OCL
with controlled NLs -- controlled English, Russian, Chinese, etc. (017)
AL
> 4. But macroassembler is still assembler, I need more. There we have objects
> that is en essence part of semantic network with defined interface. (018)
That introduces some built-in ontology to the language. That is good,
but there is no single ontology that is ideal for all purposes. I
would make these additions optional, but we can recommend some basic
ontology that is widely used. For example, numbers, time, space, etc. (019)
AL
> 5. We need language for mapping all alien's visions of the world to our's.
> I believe that this mapping should be in HOL only and therefore better
> performed in Turing Complete language. (020)
I agree with the principle. But many details need to be discussed. (021)
AL
> 6. I have as a good example of architectural language not SysML but
> ArchiMate (http://www.opengroup.org/archimate/doc/ts_archimate/ -- this
> is 1.0 specification, now we have 2.0 that is even better... (022)
That's a good example, but many details need to be discussed further. (023)
AL
> 7. Then we can have a bunch of engineering DSL languages eco-system by
> supporting translation of... (024)
That's a good list. (025)
The points I would emphasize are (1) formal semantics defined in a way
that is independent of any concrete syntax; (2) an open-ended variety
of graphic, linear, and/or controlled NL notations that are specified
by a formal mapping to the semantics; (3) the option of specializing
any notation(s) with appropriate built-in ontology; and (4) metalevel
notations for defining and relating all these languages. (026)
Following is a note I wrote to another email list. It adds some detail
about UML and other issues related to the Semantic Web. (027)
John (028)
-------- Original Message --------
Subject: Re: [CG:] Semantic Systems
Date: Sun, 04 Mar 2012 09:11:26 -0500
From: John F. Sowa <sowa@xxxxxxxxxxx>
To: cg@xxxxxxxxxxxxxxxxxxxx (029)
Danny, (030)
I want to emphasize that I'm not arguing against using the Semantic Web
notations and tools when they are useful. I also realize that there
are much more readable (and efficient) notations for RDF than RDF/XML. (031)
> I'm not sure there can be a Holy Grail, but (if you'll excuse my wooly
> logic) I agree 100% with the need for interoperability. (032)
That is the most important point I wanted to get across. (033)
> But it is a layered system - if you don't like OWL, don't use it,
> build on lower layers. (Q. Why would you want to build on any of this?
> A. to take advantage of the Web)
> RDF itself (triples) is little more than a syntax for expressing
> predicates (triples!) with URI-based naming. It doesn't really say
> anything about how those predicates should be interpreted. (034)
I also agree with those points. But my major complaint is that the
W3C started by ignoring everything else. When they developed those
notations, nearly every commercial web site was built around a
relational DB. The big guys used Oracle or DB2, and the smaller
sites used LAMP (Linux, Apache, MySQL, and Perl, Python or PHP).
In fact, they still do. (035)
If the W3C had paid equal attention to both kinds of data, they could
have promoted interoperability from Day 1. (036)
> I've not used them myself, but there are mappings between OWL and UML
> and some recent publications describe mappings between between
> SQL-based relational DBs and RDF [1]. (037)
Re OWL and UML: If you look at the OWL ontologies published on the
WWW, the overwhelming majority of them don't use anything beyond
what you can represent with UML diagrams. If the W3C had adopted
UML as their primary notation for humans, the SW would have been
fully integrated with mainstream IT from the beginning. In fact,
both sides would have benefited enormously -- and they still can. (038)
As for RDBs, Ted Codd was proposing a type system for SQL back in
the 1970s. The VLDB conferences proposed many very good upgrades
for SQL, but ISO faced two major obstacles: Oracle and IBM. (039)
The W3C had a chance to develop a good type system that could
be used with *both* SQL and the web data. Instead, their notion
of interoperability is to migrate RDBs to RDF. (040)
But people *never* migrate old software. They run the old software
forever, and they want to run it concurrently with the new software.
Note that you can still run VAX emulators on Windows, Linux, and Mac.
You can even run the PDP1 emulator on your VAX emulator on any of
the above. But it's more efficient to use the Java emulator for
the PDP1. (People do that to play the old Spacewar game.) (041)
> I agree 100% with your implication that there is a lot of untapped
> potential around Semantic Systems
> For example trust and provenance are still fairly open questions, and
> while the W3C's recent provenance data model [4] makes a lot of sense,
> my gut feeling is that to use this kind of thing in practice requires
> some input from those /other/ Semantic Systems. (042)
On that point, we completely agree. (043)
John
_______________________________________________________________________ (044)
Copy of the URLs from Danny's previous note: (045)
[1] http://www.w3.org/2001/sw/rdb2rdf/
[2] http://www.w3.org/DesignIssues/LinkedData.html
[3] http://richard.cyganiak.de/2007/10/lod/
[4] http://www.w3.org/TR/prov-dm/ (046)
_________________________________________________________________
Msg Archives: http://ontolog.cim3.net/forum/ontology-summit/
Subscribe/Config: http://ontolog.cim3.net/mailman/listinfo/ontology-summit/
Unsubscribe: mailto:ontology-summit-leave@xxxxxxxxxxxxxxxx
Community Files: http://ontolog.cim3.net/file/work/OntologySummit2012/
Community Wiki: http://ontolog.cim3.net/cgi-bin/wiki.pl?OntologySummit2012
Community Portal: http://ontolog.cim3.net/wiki/ (047)
|