[Top] [All Lists]

Re: [ontolog-forum] English number of words/concepts that cannot be comp

To: ontolog-forum@xxxxxxxxxxxxxxxx
Cc: Hassan Aït-Kaci <hak@xxxxxxx>
From: John F Sowa <sowa@xxxxxxxxxxx>
Date: Tue, 06 May 2014 14:24:09 -0400
Message-id: <536928C9.80801@xxxxxxxxxxx>
Pat C, Ed, and Kingsley,    (01)

For a useful strategy, we need to balance science (logic, linguistics),
engineering (what works), and sociology (what people actually do).    (02)

> I agree with everything John said [in the previous note]...
> My suspicion is that it will play out to some extent the way people
> learn non-native English to communicate in international conferences.
> Some do it very well, and are clear and precise and comprehensible.
> Others less so.    (03)

I am delighted that we can agree on the basic strategy for designing
useful tools that people can actually use.  I also agree that we'll
learn a lot about science, technology, and human factors as we proceed.    (04)

> The problem with controlled NLs is that they define the meanings
> of words, but they aren't very good at defining the meanings of
> constructs and sentences, unless they are extraordinarily restrictive.
> That is where ontologies shine.    (05)

Five points (the last four of which I believe are obvious):    (06)

  1. I thought that you viewed STE in a positive light. That's why I
     considered it a useful starting point.    (07)

  2. There is a continuum from unrestricted NLs, Schema.org, CNLs like
     STE, CNLs like SBVR, CNLs that map to a formal logic, and the many
     -- generally unpopular and largely ignored -- formal logics.    (08)

  3. Engineers and systems analysts have been using unrestricted NLs
     supplemented with diagrams for years -- and they still are.
     When they are gifted with an ontology stated in any formal
     logic, they *ignore* the logic and read only the comments.    (09)

  4. They aren't going to make a disruptive jump from NLs + diagrams
     to formal ontologies.  But they can be persuaded to move along
     the continuum -- provided that they are given suitable tools.    (010)

  5. If you want people to be virtuous (any way that you define virtue),
     you need to make virtue the path of least resistance.    (011)

If you disagree with #2, #3, #4, or #5, please cite the evidence.    (012)

> UML tools are used for many purposes... I would bet that
> more than half of those are actually inconsistent with fUML...
> A UML model of itself does not solve any application  problem, and
> most users do not model their implementation in sufficient detail
> for the tool to generate a complete "solution".    (013)

I agree with all those points (including those I replaced with "...").
That's why I said "such as UML", emphasized the word "supplemented",
and said that we need a methodology with suitable tools that support
the steps along the continuum.    (014)

> By comparison, an OWL model can be fed directly to an OWL reasoning
> engine that can then solve some set of application problems.    (015)

I don't disagree.  But I would emphasize the following points:    (016)

  1. OWL is only one among a huge number of logics and reasoning
     engines -- many of which are far more efficient and powerful.
     See below for an excerpt from a previous note.    (017)

  2. As you said, OWL only solves "some" of the problems.  You
     still need a lot of *supplements* to specify any practical
     application.    (018)

  3. UML can be and has been used to express the most widely used
     parts of OWL.  But UML also has diagrams that represent much
     more of logic than OWL.  UML also has the widely ignored OCL,
     which can be replaced by more readable rules stated in a CNL.    (019)

> the skills required to produce an OWL model that solves a problem are
> different from the skills required to use Java to solve the problem.    (020)

Certainly.  In fact, the skills required to produce a good UML model
are a *superset* of the skills required to produce a good OWL model.
And the tools for developing UML models are more widely available,
better understood, and much, much more widely used for mainstream IT.    (021)

> In time, the AI technologies will eliminate a lot of applications
> of Java and its relatives, and ontology development skills will
> become part of the mainstream.    (022)

> Yes!    (023)

I agree.  But my experience at IBM convinced me of the need for
a smooth migration path for any new technology.  With it, a mediocre
design can succeed (eg, Windows or Intel X86).  Without it, superior
designs are dead on arrival (eg, OS/2 or DEC Alpha). A mediocre design
without a migration path doesn't have a chance (eg, OWL).    (024)

>> But at this time, it [OWL? ontology?] is still a very small community.
>> It will grow.    (025)

> Yes, and there's more to come.    (026)

For OWL?  Or for ontology?  I have more faith in Google's R & D than
in W3C recommendations.  Google knows the importance of migration paths.
They adapted existing technology for AJAX, and it grew explosively.
They saw Linux and Firefox, and built Android and Chrome on top of them.
They saw RDF-XML and replaced the notation with JSON + RDFa.  They saw
OWL, and replaced it with Schema.org.    (027)

And note that Guha, who had been the associate director of Cyc,
the chief designer of RDF, and a developer of Schema.org at Google,
has not abandoned logic or ontology.  But he's not a fan of OWL.    (028)

I sometimes agree and sometimes disagree with Guha. But I believe that
his (or at least Google's) migration path is far superior to OWL's.    (029)

___________________________________________________________________    (030)

More recently, Amir and Aït-Kaci (2013) compared the CEDAR system
to six OWL-based reasoners:  Fact++, HermiT, Pellet, TrOWL, RacerPro,
and SnoRocket.  They compared them on four taxonomies that ranged in
size from 111,559 sorts or classes (Wikipedia) to 903,617 sorts (NCBI).    (031)

For classification, CEDAR was among the three fastest for all four
taxonomies. On the Wikipedia taxonomy, it was five times faster than
the second best (Fact++). For querying, CEDAR beat all the others by
several orders of magnitude. The query time is the most important,
since a classified CEDAR taxonomy can be saved and reused. CEDAR
also detects cycles in the taxonomy, which are a serious source of
inconsistencies.    (032)

Amir, Samir, & Hassan Aït-Kaci (2013) Fast taxonomic reasoning based
on lattice operations, CEDAR Technical Report No. 3, LIRIS-UFR
d’Informatique. http://cedar.liris.cnrs.fr/papers/ctr3.pdf    (033)

For related issues, see the three slide presentations by Hassan A-K:    (034)

Is it possible to make the Semantic Web a reality?
http://cedar.liris.cnrs.fr/papers/intis.pdf    (035)

Reasoning and the Semantic web,
http://cedar.liris.cnrs.fr/papers/ontoforum.pdf    (036)

Empirical study of high-performance triple stores,
http://cedar.liris.cnrs.fr/papers/Presentation_Cedar-PetaSky-LIP_ENS-Web-Site.pdf    (037)

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    (038)

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