W3C home > Mailing lists > Public > public-semweb-lifesci@w3.org > February 2010

Re: When does a document acquire (web) semantics?

From: John Madden <john.madden@duke.edu>
Date: Tue, 2 Feb 2010 08:25:57 -0500
Cc: w3c semweb HCLS <public-semweb-lifesci@w3.org>
Message-Id: <5AA5FEA4-168F-48CA-A307-7398A5D46BCF@duke.edu>
To: Pat Hayes <phayes@ihmc.us>
>> Is this a possible scenario? Where does it fail? Is it that the SemWeb doesn't support any notion of an "official" graph? Is it that there is no such thing as an "official graph" at all (on the sem web or anywhere else)?
> It doesn't, and there isn't. The SWeb position on official is exactly the same as the Web position, which might be summed up in the phrase Caveat Lector. Publication is easy and free and unfettered, and requires no imprimateur or legitimacy. So, its up the reader of what is published, to decide whether or not to accept it, whether or not it is trustworthy, etc.. This is why the top level of TimBL's layer cake is labelled "trust" rather than, say, "authority" or "legitimacy". And this is, of course, much like publication elsewhere, at least in the free world.

Pat, thanks. That's what I thought, and in fact that's the way I like it. It does raise an interesting question in the medical realm. Usually doctors don't express their thoughts in RDF; they usually write in (some domain-specific dialect of) English. They publish the resulting documents, giving them a signature. 

The resulting English language document *does* have a special status. For example, some doctor (#2) takes an action that he/she claims to base on some assertion in a document authored by doctor #1. Something bad happens. Doctor #2 now claims the responsibility lies with doctor #1's report, which was (claimed to be) inaccurate (contained an objective falsehood). Doctor #1 at this point would very likely produce the original report text, and ask if doctor #2 actually read it. Maybe they'd argue about whether #2's "reading" of the document is reasonable or nor.

That's all pretty routine. It gets interesting when in addition to the original English language document, doctor #1 also publishes some rendering of that document in RDF. What exactly is the relation between the RDF rendering and the original document?

I think the best way to think of the RDF -- any RDF, including the author's own RDF rendition of his very own document -- is that it is an interpretation of the original. And no interpretation is any more privileged than any other. That includes the author's own. So, like you say, caveat lector.

This makes it rather difficult to use RDF in clinical care. With English language documents, we reject those that are not signed and original (or faithful copies of the signed original) for purposes of clinical care. But in the SW world, there is no special status. You can't sign it. You can't guarantee that your triples will stick together, or be picked apart.  You can't even expect to know where a particular triple came from (who was the original assertor of the triple).

> One of the motivations for the named graph proposal was to provide exactly the kind of authoritative warrant of assertion that is discussed in the above sidebar, by the way. [1]. Y'all might want to check it out, I think it was quite ingenious. BUt no doubt too complicated for immediate adoption by the Sweb community in its current incarnation.

I think the named graphs proposal was a great piece of work, speaks directly to this, and I wish it had more uptake. It solves a huge piece of this puzzle.

There still remains the issue of what the relation is between the named graph and the original (English) document is, but at a minimum it allows some graphs to be marked with a special status.

In the interim, the closest we can come to named graphs currently seems to be the RDF document as a unit of communication. Most of the work in HCLS has so far has focused on the benefits you can get from aggregation, i.e. from specifically treating RDF documents as having no distinct identity one from another. For the clinical (as opposed to the research) use case, we do need to start thinking about ways to use RDF documents that preserve their identify qua documents.

Received on Tuesday, 2 February 2010 13:26:31 UTC

This archive was generated by hypermail 2.3.1 : Wednesday, 7 January 2015 14:52:42 UTC