- From: Hugh Glaser <hg@ecs.soton.ac.uk>
- Date: Mon, 5 Jul 2010 22:40:00 +0000
- To: Sampo Syreeni <decoy@iki.fi>
- CC: David Booth <david@dbooth.org>, Kingsley Idehen <kidehen@openlinksw.com>, "nathan@webr3.org" <nathan@webr3.org>, Pat Hayes <phayes@ihmc.us>, Toby Inkster <tai@g5n.co.uk>, Dan Brickley <danbri@danbri.org>, Linked Data community <public-lod@w3.org>, Semantic Web <semantic-web@w3.org>
Hi Sampo. I venture in again... I have much enjoyed the interchanges, and they have illuminated a number of cultural differences for me, which have helped me understand why some people have disagree with things that seem clear to me. A particular problem in this realm has been characterised as S-P-O v. O-R-O and I suspect that this reflects a Semantic Web/Linked Data cultural difference, although the alignment will not be perfect. I see I am clearly in the latter camp. Some responses below. On 05/07/2010 19:36, "Sampo Syreeni" <decoy@iki.fi> wrote: > On 2010-06-30, Hugh Glaser wrote: > >> RDF permits anyone to say anything about anything . . . except a >> literal if it is the subject of the property you want to use for the >> description. > > The way I see it, the main reason for this restriction is that the data > is supposed to be machine processable. Literals rarely are, especially > in their original, plain form. I mean, suppose we allowed literals as > subject, predicate and object. What does it really mean if I say > "Sally"@en "likes"@en "Mike"@en? > > I'd argue pretty much nothing processable. That's because literals tack > on an arbitrary, limited number of type specifiers (type and perhaps > language) to textual data, and neglect everything beyond that. That is > not how full disambiguation is done; it's how a human processor > *minimally* disambiguates a piece of text, without making it > unambiguous. > > With Schema derived or otherwise strictly derived types, the level of > disambiguation can be the same as or even better than with URI's, true. > But then that goes the other way around, too: URI's could take the place > of any such precise type. Beyond that, all that literals do is invite > people to import more ambiguity into the RDF/SemWeb framework. > > So, better to limit that to the object, in case we just *have* to have > it somewhere. (I'd rather do without entirely.) You see this as a problem of having a literal in the "subject" position. I might equally decide it is a problem with having literal in the "object" position. Literals are literals wherever they appear - they have no deeper semantics, and they certainly do not identify anything other than the literal that they are, if that makes sense. > >> So I can say: foo:booth isNamed "David Booth" But of course I can't >> say: "David Booth" isNameOf foo:booth > > You can say the same thing, as you pointed out. So you're aiming at > grammatical symmetry in excess of expressive capability. Why? There is > definite value in making the relation asymmetric: that way you can be > surer that what is being talked about stays...the subject. It's not by > any means sure that that is really going to be useful, no. Ah, perhaps the nub. The "subject" is no more the thing "being talked about" than the "object". I am not asking for symmetry of the grammar, if I understand what you mean. I am asking for the freedom to express the statements I want in the way I want, so that I can query the way I want. At the risk of repeating myself: If someone wants to say "666" foo:isTheNumberOf bar:theBeast and I have to tell them (as I do) ah, you can't say that, you need to introduce a resource numbers:666 rdfs:label "666". ... or bar:theBeast foo:hasNumber "666" I actually feel pretty stupid, having told them that RDF represents relations in a natural and basic way. In fact, I always feel a bit embarrassed when I get to the bit in my slides that shows there are two sorts of triples, as I have just said that the triples are just a directed graph. > > But at the same time it's perfectly sure that you would have to start > employing triples with both the subject and the object a literal, before > the current model can constrain you semantically. That'd then be pretty > extreme: the precise semantics of literals are tricky enough as they > stand. Pretty much the only genuine use cases I can come up with > off-hand are explicit unit conversions and label translations -- and > then anything that goes that far should probably get URI's and/or > epi-RDF conversion code in the first place. After all, both scenarios > also call for context, which might have to be disambiguated beyond > mere lexical form and type. (E.g. homologues or units with identical > dimensions but different usage.) I think you might be sort of agreeing here, but I don't understand all this discussion of context that has been going on. People seem to be somehow assuming that a literal in position one implies more "context" than one in position three, which seems strange to me. I guess that is because I am O-R-O, not S-P-O. All that being said, I am still worried about the costs of any change, compared to value returned. Although some recent posts have suggested that it is not such a bad thing. Best Hugh
Received on Monday, 5 July 2010 22:41:43 UTC