- From: Dave Reynolds <der@hplb.hpl.hp.com>
- Date: Thu, 21 May 2009 18:44:08 +0100
- To: Pat Hayes <phayes@ihmc.us>
- CC: Axel Polleres <axel.polleres@deri.org>, Boris Motik <boris.motik@comlab.ox.ac.uk>, "Eric Prud'hommeaux" <eric@w3.org>, Andy Seaborne <andy.seaborne@hp.com>, Alan Ruttenberg <alanruttenberg@gmail.com>, public-rdf-text@w3.org, Sandro Hawke <sandro@w3.org>
Pat Hayes wrote: > > On May 20, 2009, at 4:17 PM, Axel Polleres wrote: >> Undoubtedly, we are facing here some fundamental problems, which seem to >> be not restricted to RDF text alone, but in general to the extensibility >> of datatypes and D-entailment. > ** If you are right, surely the issues should be being followed by a > wider audience than this small group? ** But it seems to me that the > main issue is why we need to be using datatypes for plain literals *at > all*. I have never seen this point made explicitly, or seen what the > arguments are. Maybe this all took place inside the OWL and RIF groups, > of course, but it would be useful to have the reasons made public, if > only to help inform the technical discussions. After all, the most > elegant solution to this whole matter, from the point of view of > interoperability, would be for both OWL 2 and RIF to allow RDF literals > as they are, without transcribing them into a different syntax, and make > appropriate use of them. They are syntactically simple, easy to > recognize, and semantically transparent. Previous SWeb standards (RDFS, > OWL, SPARQL) have all accommodated to them without breaking. What has > changed? Why do we need to be having this conversation? I have been told > by various RIF and OWL2 WG members that the RDF literal design is > "broken", "a mess", "a bad design", etc.., but nobody has explained what > actual problems it creates, other than a failure to be theoretically > elegant. At the risk of confusing things further ... I don't believe RIF at least is trying to "fix" RDF lang-tagged literals. Recall that RIF is not an RDF application, it is a standard for rules interchange and does not require the data such rules work on to be in RDF. From the point of view of compatibility with the rules languages that are to be interchanged the easiest design was to have all literal constants have an associated datatype. So rif:text was something purely internal to RIF. Now RIF does include a specification (SWC) for how RIF/RDF combinations are to be interpreted. This identifies a correspondence between RDF plain literals with lang tags and RIF constants of type rif:text (as was). This seems to me harmless. The datatype is something internal to RIF translators, the datatypes don't leak into the RDF. There was no notion that rif:text should be a datatype that RDF processors would ever see or need to care about. The problems arise with the entirely well-intentioned unification of rif:text with OWL 2's similar proposed datatype and its elevation to a standalone spec. That leaves the standalone rdf:text spec trying to draw a correspondence to RDF lang-tagged plain literals outside the scope of a framework like RIF SWC which limits where and how that correspondence applies. Dave
Received on Thursday, 21 May 2009 17:45:08 UTC