W3C home > Mailing lists > Public > public-lod@w3.org > June 2011

Re: Squaring the HTTP-range-14 circle [was Re: Schema.org in RDF ...]

From: Pat Hayes <phayes@ihmc.us>
Date: Mon, 13 Jun 2011 20:33:47 -0700
Cc: Danny Ayers <danny.ayers@gmail.com>, Richard Cyganiak <richard@cyganiak.de>, Alan Ruttenberg <alanruttenberg@gmail.com>, Linked Data community <public-lod@w3.org>, Michael Hausenblas <michael.hausenblas@deri.org>
Message-Id: <08D1A172-4291-4502-AAB8-4F2BA714E724@ihmc.us>
To: William Waites <ww@styx.org>

On Jun 13, 2011, at 1:51 PM, William Waites wrote:

> * [2011-06-12 22:52:18 -0700] Pat Hayes <phayes@ihmc.us> écrit:
> ] OK, I am now completely and utterly lost. I have no idea what you 
> ] are saying or how any of it is relevant to the http-range-14 issue.
> ] Want to try running it past me again? Bear in mind that I do not
> ] accept your claim that a description of something is in any useful
> ] sense isomorphic to the thing it describes. As in, some RDF describing,
> ] say, the Eiffel tower is not in any way isomorphic to the actual
> ] tower. (I also do not understand why you think this claim matters,
> ] by the way.) 
> So in the previous email, Danny used the important word - relevant.
> Let's unpack that a little bit. Suppose we have no range-14 and all
> these RDF statements out there are all mixed up about what they refer
> to. Well, not completely mixed up. They're kind of clumped together,
> web pages and the things they are about tend to get confused but
> probably the chain of inferences that lead you to believe that the
> Eiffel tower is a dog is pretty unlikely.
> So there is some relationship between a description of the Eiffel
> tower and the tower itself. The relationship is akin to similarity in
> a very specific way - they are similar enough that someone thought it
> made sense to write down that the tower was 356m tall.

What has that got to do with the tower being similar to its description? 

> Unfortunately
> they got confused and wrote down that the web page was 356m tall. No
> matter, they are still different enough in the relevant ways that
> anyone interested in heights on the order of hundreds of meters is
> unlikely to be confused.

First, you seem to be assuming here that the tower and its description are NOT similar, contrary to what you said earlier and Danny seems to be insisting upon. Second, this hypothetical person is, we both agree, confused. They made a mistake, what they said was wrong. Correct? I ask, because many people seem to want to say that they were NOT confused or wrong, just kind of less correct than if they used the right URI. Third, and most important, anyone interested is unlikely to be confused, yes indeed. But any piece of software or inference engine is not unlikely to be confused. In fact, it is virtually guaranteed to be in the position of generating absolute nonsense.  If all the inference software was as smart as the average ten-year-old human, we wouldn't even need the semantic web because the software would be able to read the text on Web pages. But it isn't, and we do (need it, that is.) 

> Same with the dog. Is the distinction between the dog and the picture
> important to me? Maybe, maybe not. It depends what I'm trying to do.
> If I want to make sure that I can recognise the doc when I meet her,
> a picture or the actual dog might do equally well.

But if you are a semantic inference engine, and you get the dog and its picture muddled, will you likely generate a lot of nonsensical assertions? Answer, Yes, you will. Which is the key point at issue here.

> So that's the thing, similar or different in the relevant respects for
> the purpose at hand.

Yeh, yeh. Contexts, local purpose, pragmatism. Now, make this happy thought cash out in an actual logic for use on the Web. Bear in mind that the very first principle of the Web is that the *publisher* of the data, who asserts these things about dogs or pictures of dogs, cannot possibly know what 'context of use' is going to be relevant to the *user* of the published content. So I say that my picture of Fido has had its rabies shots, and what will you make of this information, for your purposes, on the other side of the planet in a foreign city years after Fido has died? And what about all the other people who will use this misinformation for their different purposes? How am I going to keep them ALL happy?

> The purpose at hand is necessary to figure out
> relevance. Just deriving all the possible things that can be entailed
> from the information you have is no good. You have to derive the
> relevant things in a particular context. You have to throw out givens
> that are irrelevant to you or that lead you to irrelevant or
> nonsensical entailments.

When you are the agent who is using this information, sure. But when you are the one publishing it or asserting it, you cannot do this. And when you are the one writing the rules to determine a globally accepted notion of entailment, you cannot do it.

> In the general case this is hard. It's not even clear if it is
> relevance understood like this is computable. The intent of the user
> is so clearly in the loop providing a reference frame for evaluating
> relevance and capturing and representing a user's intent is not
> something we have a good way of doing apart from hand-crafting
> interactions.
> Is it doable in simple cases (with rules programmed by humans) like
> figuring out the foaf:knows graph where people and their homepages
> can just be merged without too many bad side-effects.
> We need a different kind of rule here - a cut rule. That says if
> some condition obtains, *remove* some statements. For example,
> remove all { ?doc a foaf:Document } before running the productive
> rules might be a common one where we know that we aren't interested
> in information resources.

Well, now you are stepping into an ocean of cans of worms. Relevance logics, paraconsistent logics, etc. ad nauseam. But I dont think its our business to even go there. The Web logics don't give instruction on how to use information rationally in the face of uncertainty. Their purpose is much less ambitious and more restricted: just give entailment conditions which are universally correct, so that *whenever* you believe (for whatever reason) the premis, you are committed to believing the conclusion. Strict classical entailment works for everyone, and its about the only thing that does. So that is what we should be capturing in RDF and OWL, etc.. So, to go back to the http-range-14 issue, what are the *universal* principles that allow everyone to make the same valid entailments involving URI retrievals? AFAIKS, Danny is saying that there aren't any (?) Which is a reasonable answer, but is rather defeatist. I think http-range-14 is more useful than this.


> Cheers,
> -w
> -- 
> William Waites                <mailto:ww@styx.org>
> http://river.styx.org/ww/        <sip:ww@styx.org>
> F4B3 39BF E775 CF42 0BAB  3DF0 BE40 A6DF B06F FD45

IHMC                                     (850)434 8903 or (650)494 3973   
40 South Alcaniz St.           (850)202 4416   office
Pensacola                            (850)202 4440   fax
FL 32502                              (850)291 0667   mobile
phayesAT-SIGNihmc.us       http://www.ihmc.us/users/phayes
Received on Tuesday, 14 June 2011 03:34:44 UTC

This archive was generated by hypermail 2.3.1 : Wednesday, 7 January 2015 15:16:14 UTC