Re: KR and RDF Redux

In connection with your remarks about RDF's (lack of) semantics, let me make a couple of remarks about KR's rich semantics.

Given the sentence
    at past { John do go to the store done }
KR's built-in "tabula rasa" context says:
    past is a time context (potentially -- I think I deleted that in a recent housecleaning spree)
    John is an entity, because only entities do actions
    go is an action
    the store is a "final characteristic" of "go" (additional context could say that it is a physical location)

Beyond that, you can add as much or as little context as you want.
e.g., you can give a genus-differentia definition which distinguishes "go" from other actions
e.g., you can uniquely identify John with a genus-differentia definition like
John isu person with   # "isu" is allowed to be replaced by "is" for backwards compatibility
    name = nnn,
    birthdate = ttt,
    mother = mmm,
    father = fff

Note: Some people (including me) talk about using Social Security Number for a definition.  There are two serious problems with that idea:
(1) SSN is not unique -- they are reused !!! -- and some people don't have one; (2)if you have a unique number, that's good, but that's not in the spirit of a genus-differentia definition.  The differentia are fundamental characteristics of John, the characteristics which make him John, and differentiate him from other people.  I'd say the ultimate differentia would be his DNA profile, which will probably be doable in the near future.  But of course a unique number is more practical, and a desirable bookkeeping tool in addition to a definition.

To reiterate, you have great flexibility in the amount of context you put into KR's knowledge base.
If you don't supply any context, KR mostly treats words as uninterpreted strings, as RDF does.
If you supply a rich context, you can do very sophisticated reasoning with KR.
============ 
Dick McCullough 
knowledge := man do identify od existent done
knowledge haspart list of proposition

  ----- Original Message ----- 
  From: Frank Manola 
  To: Richard H. McCullough 
  Cc: Jon Hanna ; www-rdf-interest@w3.org 
  Sent: Friday, November 22, 2002 2:21 PM
  Subject: Re: KR and RDF Redux



  Richard H. McCullough wrote:

  > Frank
  > 
  > You raised a lot of questions, and I don't think it would be helpful for 
  > me to try to address every one individually.  Although I'd be  happy to 
  > do that off-list if you like.


  Maybe at some point, but no time right now (deadlines....)


  > 
  > In the meantime, I'm going to make a general remarks.  
  > 
  > 1. You had a long list of concepts with little structure.  If you 
  > continue like that you're going to end up with 100,000 properties to 
  > describe 10,000 concepts.
  > 


  Sure, but that depends on the extent to which you want to decompose 
  things.  I think, though, you can see from the example I gave that all 
  this is doing is taking something like

  ex:John ex:wentTo ex:theStore

  which bundles a whole lot of interpretation into the parts of one 
  triple, and decomposing the ideas into simpler ones (like action verbs 
  with agents who are the performers, and so on).  But you still have to 
  read meaning into all that stuff;  RDF doesn't provide that meaning. 
  All RDF can express is relationships between things.  So all that triple 
  means as far as RDF is concerned is that there are two things, ex:John 
  and ex:theStore, and there's some relationship called ex:wentTo between 
  them.  Strictly speaking, you can't say much about the nature of that 
  relationship either.  You can say (in RDF Schema) what its domain and 
  range are supposed to be, and you can write triples using other 
  properties that further describe the relationship itself, e.g.

  ex:wentTo rdf:type ex:verb

  but again, all you're expressing, as far as RDF is concerned, is that 
  the relationship ex:wentTo, considered now as a thing, has a 
  relationship rdf:type with some thing ex:verb.  RDF per se doesn't 
  capture what "verbness" is;  just what you claim its relationships with 
  other things are.  If I understand your KR example correctly, KR has 
  much richer built-in semantics (NB:  I was assuming in your earlier 
  message that "KR" referred to "knowledge representation" in general, not 
  to a specific notation.  Clearly I missed some earlier context).


  >  
  > 
  > 2. I checked my bookshelf, and I don't own Sowa's "Knowledge 
  > Representation"; I suppose that's a recent book?  I don't think I would 
  > find any surprises in his book because I have read a number of his 
  > papers, and a book which he edited.


  Copyright 2000.  It does cover much of his earlier work.


  > 
  >  
  > 
  > 3. As for the example:
  > 
  > English:     John went to the store
  > 
  > RDFS:      <xx:John> <xx:wentTo> <xx:TheStore>
  > KR:           at past { John do go to the store done }
  > 
  >  
  > 
  > It was part of my design philosophy for "simplifying English" to avoid 
  > changing names to describe verb tenses, plurals, etc.
  > 


  One of the things Sowa talks about a lot is the relationship of logic to 
  various "controlled English" approaches.  The comparison would be 
  interesting (but not now!).

  --Frank



  -- 
  Frank Manola                   The MITRE Corporation
  202 Burlington Road, MS A345   Bedford, MA 01730-1420
  mailto:fmanola@mitre.org       voice: 781-271-8147   FAX: 781-271-875