Dear John, Richard, Azamat, Doug, et al,
(BI can understand if you don't find the patent
(Bexample comfortable, and want to try another area
(Bas our common focus. But I also have trouble with
(Bthe VivoMind example because it is closed - you
(Bhave code and methods that have not been well
(Bdefined and described for public consumption. I
(Bsuggested patents because there is such a regular
(Bform to them, and because the database is open to
(Bpublic query and comparison.
(BWe need a SMALL example, a SIMPLE example, and a
(BWELL DOCUMENTED world, IMHO, to make progress on
(BDoug's already well developed nucleus of
(Bmicrotheory, which is a great starting point.
(BWhat can we choose that will be acceptable to all
(Bof us, small enough to make progress with, and yet
(Bable to clear up our early diversity of
(Bviewpoints? I am open to suggestions, but so far
(Bnone of us have come up with the right focus yet.
(BAt one point, you wanted to use biosemiotics and
(Bapply Peirce's thoughts to the problem. So I
(Btried to respond with Use Case 1, but that hasn't
(Bstirred up progressive discussions either. So I
(Btried Use Case 2 which was more abstract. We
(Bcould map Use Case 2 onto Use Case 1 and perhaps
(Bdeal with the self-interest versus interest issue.
(BWould that help us meet on a common ground for
(Bclearing up the self-interest vs interest issues?
(BRich AT EnglishLogicKernel DOT com
(B9 4 9 \ 5 2 5 - 5 7 1 2
(BBehalf Of John F. Sowa
(BSent: Sunday, August 28, 2011 12:48 PM
(BSubject: [ontolog-forum] Semantics of Natural
(BDoug, Rich, and Azamat,
(BI changed the subject line because there are some
(Bdifferences in our assumptions and ways of
(Bworking. We're going
(Boff in totally different directions that are
(Bunlikely to converge.
(BBefore going further, we should review where we
(Bplan to go and how.
(BDoug worked at Cyc, and he understands the Cyc
(Bbase, and methods of reasoning. He sat down and
(Bwrote a precise,
(Bwell-defined microtheory based on the Cyc
(Bontology. It defines
(Ba set of terms that we have been discussing in a
(Bway that a system
(Bsuch as Cyc could use to interpret sentences about
(Bdraw inferences, and answer questions.
(BRich started the thread for a self-interest
(Bontology because he
(Bwanted to address questions about how governments
(Bwork: How are
(Bthe laws, policies, and regulations of a
(Bgovernment related to
(Bthe needs and interests of the people in a
(B> It seems to me that Doug$B!G(Bs initial ontology is
(Bat the Theory level...
(B> Perhaps instead we should... try to make
(Bprogress elaborating Doug's
(B> formulation by experiment, observation or
(B> but in a more focused manner.
(B> I suggest... that we consider US patent
(Bspecifications as the narrow
(B> class of concise situation descriptions, problem
(B> that situation, and claimed embodiments of
(BAzamat has a much grander goal of trying to
(Bspecify a global ontology
(Bof everything. He wants something more
(Bprescriptive than descriptive:
(B> Nowadays, the scope of human and national
(Binterest is formed by
(B> domineering politics, ideology, or commercial
(B> Now a sensible ontology of self-interest is
(Bsupposed to raise the
(B> eco-awareness, eco-interest as well, to help
(Bpeople see their
(B> responsibility for the environment...
(BI sympathize with all three of these goals. But
(BI'd like to describe
(Bwhat we've been doing at our VivoMind company. We
(Bhave some good
(Btools, but we only have a very small group, and we
(Bdon't have any
(Bresearch funding. We have to work on projects our
(Bclients will pay
(Bus to do, and we have to deliver the kinds of
(Bresults they want with
(Bwhatever budget they're willing to approve. But
(Bat the same time,
(Bwe're trying to develop our technology in ways
(Bthat can solve very
(Bchallenging problems of natural language
(BI agree with Rich "that Doug$B!G(Bs initial ontology
(Bis at the theory level."
(BIn fact, that has been one of the complaints about
(BCyc that I discussed
(Bwith Lenat since the 1990s. But I strongly
(Bdisagree with the word
(B'narrow' in Rich's phrase: "the narrow class of
(Bdescriptions, problem statements within that
(Bsituation, and claimed
(Bembodiments of solutions."
(BI realize that Rich has done a lot of work with
(Bpatents, but there's
(Ba huge difference between the narrow conventions
(Bfor stating the
(Bpatent application and the unbounded subject
(Bmatter of all the things
(Bthat could be patented. Furthermore, the patent
(Blawyers try to make
(Bthe invention sound novel by deliberately using
(Bterminology that is
(Bbased on standard terms for the context, but
(Bdifferent in detail
(Bfrom anything that anyone else has done.
(BIt's conceivable that an ontology of self-interest
(Bthe inventor's self interest to the parts of a
(BBut that would also be "at the theory level", not
(Bat the level
(Bthat any client is likely to pay somebody to do.
(BBefore doing any work on patents, I want to see an
(Bthat a real customer with real money is willing to
(Bto solve. Give us some patents or patent
(Bapplications and a
(Bspecific problem that some paying client wants to
(BAs for the environment, I would strongly support
(Bany work that could
(Bhelp preserve the environment, wildlife, etc. But
(Ban ontology has
(Bto clarify the subject matter. It should not have
(Bvalue judgments about what should or should not be
(BFinally, I'd like to mention something about the
(Blevel of ontology
(Bthat we (at VivoMind) have found most useful. For
(Bthe applications we have worked on, see the
(BFor an excerpt from the kind of "English", see
(Bslide 27. Trying to
(Btranslate that text directly to CycL (or any other
(Bkind of logic)
(Bwould be impossible. Instead, note the method
(Boutlined in slide 26:
(B> Much easier task:
(B> $B!|(B Translate the COBOL programs to conceptual
(B> $B!|(B Use the conceptual graphs from COBOL to
(Binterpret the English.
(B> $B!|(B Use the analogy engine to compare the graphs
(Bderived from COBOL
(B> to the graphs derived from English.
(B> $B!|(B Record the similarities and discrepancies.
(BThe Intellitex parser used a general ontology with
(Bvery few axioms.
(BBut the details needed for the application came
(Bthe data structures and definitions in COBOL,
(Badding all the
(Bnames of files and programs to the English
(Btranslating that information to conceptual graphs
(BThen those CGs served as the semantic foundation
(Bthe English sentences. Any sentences that did not
(Brefer to anything
(Bderived from the COBOL programs were ignored as
(BNote that this application did not require any
(Bor any detailed ontology other than a simple
(Bhierarchy of terms.
(BBut it did require a very large and detailed
(BFortunately, that ontology could be automatically
(Ba formal language (COBOL).
(BI won't claim that this is the only way to do
(Bbut it worked very well. It shows how a very
(Bontology can be used to solve a complex problem
(Bany axioms from an upper-level or mid-level
(Bontology. It did,
(Bhowever, require the kind of lexical information
(Bthat could be
(Bderived from WordNet or something similar.
(BI'll admit that more details at the upper and mid
(Boften be useful. But it is also possible to
(Bderive much of that
(Binformation automatically by analyzing documents
(BThe application in slides 33 to 41 is an example
(Bthat shows how.
(BSlide 34 shows the source material, which included
(Bsome of which were research reports and others
(Bfrom a textbook on geology. Before answering any
(Bthe VivoMind system translated all the documents
(Bgraphs and indexed them with the Cognitive Memory
(BSlide 40 shows how the query (a paragraph written
(Bby a professional
(Bgeologist) was related to the answer by using
(BChapters 44 and 45 of the textbook.
(BThe answer to the geologist's query was derived
(Bfrom a research
(Breport (McCaffrey and Kneller 2001), but it was
(Bto match the sentences in the query directly to
(BHowever, Chapter 44 contained information about
(B"lowstand fan", "passive margin", and "turbiditic
(BChapter 45 contributed information about three
(B"narrow feeder corridors", "stratigraphic onlap",
(BSlide 41 summarizes the method:
(B> Emergent Knowledge
(B> When reading the 79 documents,
(B> $B!|(B VLP translates the sentences and paragraphs
(B> $B!|(B But it does not do any further analysis of
(B> When a geologist asks a question,
(B> $B!|(B The VivoMind system may find related phrases
(Bin many sources.
(B> $B!|(B To connect those phrases, it may need to do
(B> $B!|(B The result is a conceptual graph that relates
(Bthe question to
(B> multiple passages in multiple sources.
(B> $B!|(B Some of those sources might contribute
(Binformation that does not
(B> have any words that came from the original
(B> $B!|(B That new CG can be used to answer further
(B> By a $B!H(BSocratic$B!I(B dialog, the geologist can lead
(Bthe system to
(B> explore novel paths and discover unexpected
(BNote that these applications show that the
(Bhighly domain-dependent information is the most
(Bthat information is usually highly voluminous. It
(Bor impossible to define all of it in advance,
(Bespecially by hand.
(BBut automated methods can often derive that kind
(Bfrom unstructured, natural language documents.
(BShared Files: http://ontolog.cim3.net/file/
(BCommunity Wiki: http://ontolog.cim3.net/wiki/
(BMessage Archives: http://ontolog.cim3.net/forum/ontolog-forum/
(BConfig Subscr: http://ontolog.cim3.net/mailman/listinfo/ontolog-forum/
(BShared Files: http://ontolog.cim3.net/file/
(BCommunity Wiki: http://ontolog.cim3.net/wiki/
(BTo join: http://ontolog.cim3.net/cgi-bin/wiki.pl?WikiHomePage#nid1J