- From: Sergey Melnik <melnik@db.stanford.edu>
- Date: Thu, 25 Oct 2001 17:54:46 -0700
- To: Pat Hayes <phayes@ai.uwf.edu>
- CC: w3c-rdfcore-wg@w3.org
Pat Hayes wrote: > > >Pat Hayes wrote: > >> ... > >> Right, but one would not use IEXT for reification. I introduced a > >> REIF mapping from the syntax into IR, but one could do it more simply > >> just by requiring IR to contain a suitable set of expressions, by > >> semantic fiat, so that REIF is defined to be the identity. > >> > >> > > >> >What about the following addition to the MT. Let a ternary relationship > >> >Reif be defined as: > >> > > >> >1) IR x IR x (IR union LV) <= Reif > >> >2) If (x,y,z) in (IR union Reif) x IR x (IR union LV union Reif), > >> > then (x,y,z) in Reif > >> >3) Reif is the smallest set with (1), (2) > >> > > >> >In other words, Reif contains all reified statements that can possibly > >> >be constructed under a given interpretation I. > >> > >> Right, that is almost the version I had at the F2F. (I used a mapping > >> from the syntax into the domain, you use a three-place predicate > >> defined recursively on the domain.) However, it was resoundingly > >> criticized as not what reification actually means, in practice. (Dan > >> C. told me that this was what reification *ought* to have been, but > >> in fact it wasn't that.) I am still waiting to discover what > >> reification actually is, in practice (as opposed to what the M&S says > >> it is.) > > > >Maybe the confusion arose because in your original version reification > >mapped directly from the syntax. I must confess that I thought (and > >possibly other F2Fers did) that your definition treated reification as > >some kind of special syntactic mapping that needs to be established in > >addition to I. > > Well, in a sense it is, though its an embedding rather than an > interpretation (mention, not use!). > > If I follow what Jeremy C. was saying in his whats-the-fuss-about > message concerning reification, then *some* kind of embedding from > the syntax seems to be essential, since the rdf:subject and > rdf:object arcs both point and refer to the very pieces of syntax > that are in the triple being reified. If we can find a way to avoid > the use/mention confusion (and Im rapidly coming to think that we > can, and it would be based on the same trick that we used to handle > literal overloading, by allowing the objects of the reification arcs > to be the nodes themselves, rather than what the node labels denote) > then this would I think be able to handle reification in just the way > that users intend it to be used. The resulting system would be a > rather unorthodox logic, and the reification relations would have a > very odd relationship to the syntax; but that is OK, since they do :-) Sorry, I'm slow here. Why do we need to pull the pants over the head? Please explain what the problem is with the following approach: Given: four statement below. s p o id rdf:subject s id rdf:predicate p id rdf:object o Possible interpretation: DD = { I(s), I(p), I(o), ... } union Reif Reif = { <I(s),I(p),I(o)>, ... } I(id) = <I(s),I(p),I(o)> IEXT(I(p)) = { <I(s),I(o)> } IEXT(I(rdf:subject)) = { < <I(s),I(p),I(o)>, I(s) > } IEXT(I(rdf:predicate)) = { < <I(s),I(p),I(o)>, I(p) > } IEXT(I(rdf:object)) = { < <I(s),I(p),I(o)>, I(o) > } If you don't want to deal with Reif and identify the thingies explicitly, the interpretation becomes: DD = { I(s), I(p), I(o), I(id), ... } IEXT(I(p)) = { <I(s),I(o)> } IEXT(I(rdf:subject)) = { < I(id), I(s) > } IEXT(I(rdf:predicate)) = { < I(id), I(p) > } IEXT(I(rdf:object)) = { < I(id), I(o) > } where I(id) is just some resource in DD. Why is some kind of "embedding from the syntax" need? What is the mention/confusion in the above interpretation? If we discard Reif, why does reification require some special treatment in MT at all? I think the above interpretation is consistent with the current MT draft. Sergey
Received on Thursday, 25 October 2001 20:28:22 UTC