- From: Hart, Lewis <lhart@grci.com>
- Date: Fri, 20 Oct 2000 09:49:40 -0400
- To: Jeff Heflin <heflin@cs.umd.edu>
- Cc: www-rdf-logic@w3.org
>-----Original Message----- >From: Jeff Heflin [mailto:heflin@cs.umd.edu] >Sent: Thursday, October 19, 2000 3:31 PM >>"Hart, Lewis" wrote: >> >> >> To me the definition of 'globally agreed to' means a specified version >> of the ontology, as given by its URL, which everyone who cares knows >> about. The kid in the Philippines could publish whatever he wants, but >> it doesn't mean anyone will use it. If an author intends to use the >> agreed to meanings, then they must reference the appropriate defining >> ontology: >> >> <rdf:RDF >> xmlns:rdf="http://www.w3.org/1999/02/22-rdf-syntax-ns#" >> xmlns:daml="http://www.w3.org/2000/08/daml-ont#" >> >> xmlns:EC="http://www.e.commerce.org/standards/ontology#" >> >> > >> <EC:Purchase_Order > >> ... blah blah blah >> </EC:Purchase_Order > >> </rdf:RDF > >> >> If a different ontology (say xmlns:EC="http://hackers.net/ontology") is >> referenced, >> then you do not necessarily know what is meant. You may or may not trust it. >> > >The problem comes not so much in interacting with a site or exchanging >messages between agents (in which case certificates and/or case-by-case >acceptance might work sufficiently), the problem comes with an >"ontology-based" search engine. Here's an overly-simplified example: > >Let's say we have the following in our E-commerce ontology: > >Product isa Thing >Computer isa Product >... > >Let's say someone has built the first quantum computer and is offering >it for sale. They need to extend the E-commerce ontology with a special >QuantumComputerOntology, which might include the following: > >QuantumComputer isa Computer > >Finally, on a web page this person now says: > >Q2000 instance-of QuantumComputer > >Let's say I'm in the market for a new computer. I might be interested in >the QuantumComputer if I knew about it, but if I didn't know about the >QuantumComputerOntology beforehand, I couldn't say that I trust this >ontology. Thus, my query about Computers will never return this instance >b/c I don't believe that "QuantumComputer isa Computer." In this case, >it would be nice if my reasoner automatically used the new ontology, so >that my query would return the Q2000 as well as more traditional >computers. Agreed. Perhaps the approach to take could be termed "optimistic authentication", assume that the ontology is a good thing until shown otherwise. The onus will be upon the consumers of the ontology to determine its goodness, as applied to their domain/problem. > >However, let's say that some hacker creates an ontology the extends the >E-commerce ontology with this beauty: > >Thing isa Computer > >Now if my reasoner automatically accepts this ontology, I'm in for a >world of trouble. It will tell me that everything is a computer! It is not the same Thing (pun intended). Remember that there are multiple namespaces involved. A more exact summary of the ontological statements above might be, for the valid case: ec:Product isa daml:Thing ec:Computer isa ec:Product qco:QuantumComputer isa ec:Computer qco:Q2000 instance-of qco:QuantumComputer and for the invalid one: bad:Thing isa ec:Computer Four different ontologies are in use here: DAML (daml:), eCommerce (ec:), the QuantumComputerOntology (qco:) and the bogus one (bad:). What was actually said is not all dmal:Things are ec:Computers but just all bad:Things are ec:Computers. Perhaps this is not even bad in itself, if all that this ontology discusses are in fact computers. > >The problem is we want different behaviors depending on our a priori >knowledge of the ontology. How do we make sure the right thing is done? >I don't want users of the brave new web to be asked every minute by >their computers if they accept John Doe's new ontology (which may be >very complicated and contain implications that they don't understand), >but I also don't want some guy selling widgets out his basement to have >to wait 3 months for some certification body to say his ontology is >trustworthy before people can even find out that he sells that kind of >widgets! That goes totally against the distributed nature of the Web! > > >Jeff Agreed. - Lewis ___________________________________________ Lewis L Hart GRC International lhart@grci.com 1900 Gallows Rd. Voice (703)506-5938 Vienna, Va 22182 Fax (703)556-4261
Received on Friday, 20 October 2000 09:49:46 UTC