W3C home > Mailing lists > Public > public-sw-meaning@w3.org > May 2004

Re: Comment on "Named Graphs, Provenance and Trust"

From: Bijan Parsia <bparsia@isr.umd.edu>
Date: Tue, 11 May 2004 09:41:09 -0400
Message-Id: <DC4A36B3-A350-11D8-84C1-0003939CCC42@isr.umd.edu>
Cc: public-sw-meaning@w3.org, "John Black" <JohnBlack@deltek.com>
To: Pat Hayes <phayes@ihmc.us>

On May 6, 2004, at 3:11 PM, Pat Hayes wrote:
[snip]
>> This is what I previously referred to as a 'stipulative definition' or
>> 'stipulative ontology'.  What I am trying to get at intuitively is the
>> ability to say, "When this term is used in this context it SHALL be
>> interpreted to mean that".  And as a crude example of its use, imagine
>> I had need to import two ontologies, one a universal business language
>> ontology of commercial transactions, and the other, Bijan and Peter's
>> alternative ontology of transactions, both of which contained some 
>> terms
>> I needed but which had an incompatible "invoice" term that caused an
>> ambiguity (inconsistency).  Now I want to be able to eliminate the
>> ambiguity by stating that within this named graph, when the term
>> "invoice" is used, it SHALL be interpreted according to the UBL 
>> ontology
>> (or vice versa).

(The above was from John Black.) Interestingly, John, this is exactly 
the kind of scenario we were interesting in making possible, even 
though the current tools (owl:imports, cut and paste, etc.) are a bit 
crude. The current canonical way to achieve this is to simply only 
include in your document the desired axioms. On some views (e.g., some 
variant's of Tim's view), it was much more difficult to exclude the URI 
owner's axioms.

However, as pat point out, you can't stop me (currently, or really, 
ever) from adding yet more axioms to your set. E.g., if I import *your* 
document, I'm almost surely going to add assertions, and I'm free to 
add axioms, even concerning your classes. Even if I just add facts 
("instances"), I can affect the *inferred* axioms (e.g., the 
classification). So the question becomes, how far reaching should your 
stipulations reach?

I *like* being able to say, of my documents, that the axioms are 
*these* ones I put in some effort to put in, and not *those* which I 
put in some effort to exclude.

> But the performative story only gives you a kind of act, not a 
> stipulation about other's acts or interpretations. So let me try to 
> rephrase this a little: suppose we say that the intuitive force of 
> such a stipulative definition is to declare that *your intention* is 
> that the terms shall have the meaning you give to them, and that any 
> usage (such as Bijan and Peter's, in the example) which clashes or is 
> incompatible with your meanings is not intended by you. This of course 
> does not *prevent* P & B from misusing your terminology (how could 
> it?) , or force anyone else to believe you rather than them (so all 
> this foolish

Foolish stupid? Foolish tactically unwise? Foolish cute?

> analogizing with tyranny, oppression and freedom is beside the point)
[snip]

I dispute that it's (necessarily) a misuse, so perhaps the analogy, 
about which I was having a few qualms, is apt.

The point of this discussion is how much de jure authority we are 
giving to URI owners. On some of the views that Peter and I opposed, 
the only sanctioned way to avoid slurping in the URI owner's URIs was 
to use only your own URIs.  Perhaps we're all past that now, but at the 
time such views were seriously expoused and argued for.

I actually don't *know* what the right default is, but I'm pretty sure 
that a link/transclusion distinction is useful wrt to URIs in RDF and 
OWL documents, just as it is useful in HTML pages. Looking at HTML, I 
tend to go with linking being the default. Given that it seems quite 
possible that the most interesting sets of axioms concerning a resource 
might *not* be published by the URI owner, I'm not convinced that 
transclusion (rather than keying to other aggregated sources) is the 
most useful mechanism. Pervasive transclusion just seems to be nuts. 
Pervasive, uspecified transclusion (oh, this agent could do SOC and 
this one might now, but who knows, it depends) is an interoperability 
problem. owl:imports gives us some sort of discipline for 
transclusions.

As for forcing vs. non forcing, either our standards mean something and 
we want to be able to mobilize arguments for this or that behavior 
based on standards compliance, or we should pack up and take a nice 
vacation. People who are irked at OWL-DL or OWL-Full features can, in 
fact, write software that processes OWL ontologies in non-compliant 
ways, but there are a lot of social pressures not to do so. 
Furthermore, there is and has been a lot of talk, much of it loose, of 
making/supporting various laws regarding semantic web content. (At 
these points, I remember: 
http://www.xml.com/pub/a/2001/01/31/politics.html. I should nag kendall 
about that. I suspect there's a good deal of update to be done.)

I had more but this is coming late enough as it is (curse grading and 
all its works).

Cheers,
Bijan Parsia.
Received on Tuesday, 11 May 2004 09:43:33 GMT

This archive was generated by hypermail 2.2.0+W3C-0.50 : Monday, 7 December 2009 10:42:16 GMT