- From: pat hayes <phayes@ihmc.us>
- Date: Tue, 13 Jan 2004 20:23:22 -0600
- To: "Peter F. Patel-Schneider" <pfps@research.bell-labs.com>
- Cc: herman.ter.horst@philips.com, jjc@hpl.hp.com, hendler@cs.umd.edu, schreiber@swi.psy.uva.nl, connolly@w3.org, sandro@w3.org, www-rdf-comments@w3.org, bwm@hplb.hpl.hp.com
>From: pat hayes <phayes@ihmc.us> >Subject: Re: RDF Semantics: corrections >Date: Tue, 13 Jan 2004 17:37:33 -0600 > >> >From: pat hayes <phayes@ihmc.us> >> >Subject: Re: RDF Semantics: corrections >> >Date: Tue, 13 Jan 2004 15:36:29 -0600 >> > >> >[...] >> > >> > >> >> For clarification, my question to you was concerned with the effects >> >> on OWL rather than on RDF as such. I was not meaning to ask your >> >> opinion on the change in a general sense: as indicated, I will take >> >> responsibility for that. I was asking you only to give an opinion on >> >> whether the suggestion made by Herman for a 'protective' modification >> >> to the *text* of the OWL document is indeed sufficient to ensure that >> >> the RDF and OWL specs, with these changes, will still be compatible. >> >> The intention is that this will make no change to OWL itself. >> > >> >Well, you are really asking for us to approve both the change to RDF and >> >related change to OWL to insulate OWL from the effect of the change to >> >RDF. From what I see about the changes I would vote against both. >> >> No, any changes or otherwise to RDF will be the result of decisions >> made by the RDF WG. What I was really asking your opinion of was the >> insulation device proposed by Herman. There are three options: >> >> 1. do nothing >> 2. this change is made to RDF and a textual >>edit in OWL insulates OWL from it >> 3. This change is made to RDF and OWL is damaged as a result >> >> I am anxious to avoid 3. I know you prefer 1., but what I was really >> asking you is whether 2 is possible, or whether the choice is between >> 1 and 3. If indeed that is the only possible choice, I will have to >> choose 1. I myself would vastly prefer 2 to 1, and I think that the >> overall design will be better and will be more generally useful with >> 2 rather than 1. So I know you would vote for 1 given your druthers: >> but can you please tell me if, given the choice between 2 and 3, >> whether 2 is even an option? > >Well, if I was voting I would vote against 2. Yes, I understand that. My question is, however, is 2 even possible? That is, I want to know if I (actually the RDF WG) am choosing between 1 and 2, or will I (etc.) be forced to choose between 1 and 3? My (our) decision depends on the answer. > >> >[...] >> > >> >> The substantive change is that, as noted above, the class extension >> >> of a datatype name is required to contain only the denotations of >> >> typed literals in the vocabulary, rather than required to *be* the >> >> value space of the datatype. This brings D-interpretations in line >> >> with the way that XML typed literals are treated in the RDF >> >> semantics, which imposes no explicit requirements on >> >> ICEXT(I(rdf:XMLLiteral)) >> >> >> >> The only example I am aware of which makes the change visible is this: >> >> >> >> ex:a ex:p "true"^^xsd:boolean . >> >> ex:a ex:p "false"^^xsd:boolean . >> >> ex:c rdf:type xsd:boolean . >> >> >> >> |=?= >> >> >> >> ex:a ex:p ex:c . >> > >> >What about >> > >> >ex:a ex:p "2"^^xsd:short . >> > >> >|=?= >> > >> >ex:a ex:p _:a . >> >_:a rdf:type xsd:int . >> > >> >I think that this is a valid {xsd:short,xsd:int}-entailment in the PR >> >version of RDF but would not be valid under the proposed change. >> >> Hmm. It is hard to tell since the XSD specs seem to be ambiguous >> about whether or not value spaces of distinct datatypes are disjoint >> or not. > >In this particular case it is even more clear than in the case of xsd:float >vs xsd:int that the values spaces are not disjoint, as xsd:short is a >restriction of xsd:int, and thus its value space is, by definition, a >subset of the value space of xsd:int. Ah, yes, I had not noticed that. OK, in that case I concede that one reasonable way to interpret the XSD spec in this case is that the value space is a true subset. However, these are the only cases, since there is an explicit prohibition (in section "4.2.1 equal") against any other equalities being considered true; and since other parts of the text of the document strongly suggest otherwise (for example the definitions of the value spaces of hexBinary and of base64Binary are both stated to be "the set of finite-length sequences of binary octets", yet these sets are required to be disjoint by section 4.2.1), the moral I draw from this text, one reinforced by subsequent discussions with members of the WG, is that it is using a notion of 'equality' and/or of 'set' which does not entirely correspond to normal mathematical usage, and in any case is certainly open to reinterpretation: so I retain my preference for adopting a conservative notion of identity when it comes to formal semantics. One merit of the proposed change is that it entirely insulates RDF D-entailment from the answers to such questions, in fact: D-validity would depend only on the lexical form of literals. I believe this is closer to the intentions of the XSD designers, since it is closer to the way that datatypes are used in programming environments, where questions about class extensions of datatypes rarely arise (and are in any case determined, when they do, by the particular language rather than by the XSD spec.) > > My best understanding of the intent of the XSD group (and >> what they were planning to make official in version 2, the last time >> I checked) is that those value spaces should be considered to be >> disjoint in the sense that their members are not identical, but that >> there is a relation called 'equality', distinct from the relation of >> identity, which holds between them. (An alternative account of the >> situation was provided by Henry Thompson, who indicated that the best >> way to think of the elements of these value spaces is as pairs >> consisting of the value plus the datatype: rather like typed >> literals, in fact.) Whether or not this 'equality' is the same as >> owl:sameAs, I leave others to decide. I tend to assume that the >> safest thing for us to do in the semantics is to go with the most >> conservative possibility. As far as the RDFS semantics is concerned, >> therefore, the value spaces of any two different XSD datatypes MAY be >> distinct disjoint sets, so your example is not a valid entailment on >> either the PR or with the proposed change. > >I totally disagree. The normative documents for XML Schema are totally >clear The wording seems unambiguous, I agree, but unfortunately it seems just as unambiguous in other cases where it clearly cannot be interpreted in this way. Hence my enduring mistrust of what exactly was intended. This discussion is slightly aside from the main point, but let me phrase it this way. Suppose some folk were having a debate about your example, using the PR RDF semantics. A says that it is valid since the class extension of xsd:short is required to be a subset of that of xsd:int. B says that doesnt follow because it is possible to read the XSD spec in such a way that if a is of A and b of B, with A and B distinct built-in datatypes, then a =/= b, so interpretations exist in which the class extensions are disjoint. I would side with B, myself; but more to the present point at issue (the proposed change), I would prefer that the debate had never been allowed to arise in the first place, and that D-entailment was insulated from such questions as whether or not facets confer datatype identity: particularly since the relevant point at issue can be expressed directly in RDFS: so, rather than arguing about what is 'true', what A and B ought to do is for each of them to be explicit about their assumptions, and then they will be able to, if not agree, at least be usefully explicit about what they are disagreeing about. If B asserts the relevant subclass relationship, then A will know what his assumptions are. They might disagree about the *need* to do this, since A could argue that it is obvious: but then, we all need to acknowledge the unfortunate necessity, when communicating with others, to be explicit about things that seem obvious to us. >, particularly in this case, that the value spaces are not disjoint. >(Note that this is a separate issue from whether a value typed as a short >would validate as an int.) > >> It would be valid if you also asserted >> >> xsd:short rdfs:subClassOf xsd:int . >> >> under either semantics. > >Well, yes, but so what? For example what about switching xsd:short and >xsd:int above? So, if you want subclass relationships to be involved in your reasoning, you need to state your assumptions explicitly. That seems to me to be an eminently appropriate methodological conclusion for RDFS. I fail to follow your point about switching. Pat PS. One point about the change which I guess I didnt make explicit is that SOME change is required: the text in the PR version is buggy, in a way that Herman identified (acting, as I understand it, on behalf of the Webont group, by checking the text extremely carefully). So 1. above should really say, rather than 'no change', 'not this change'. > >> Pat >> -- >> --------------------------------------------------------------------- >> IHMC (850)434 8903 or (650)494 3973 home >> 40 South Alcaniz St. (850)202 4416 office >> Pensacola (850)202 4440 fax >> FL 32501 (850)291 0667 cell >> phayes@ihmc.us http://www.ihmc.us/users/phayes > >Peter F. Patel-Schneider >Bell Labs Research -- --------------------------------------------------------------------- IHMC (850)434 8903 or (650)494 3973 home 40 South Alcaniz St. (850)202 4416 office Pensacola (850)202 4440 fax FL 32501 (850)291 0667 cell phayes@ihmc.us http://www.ihmc.us/users/phayes
Received on Tuesday, 13 January 2004 21:23:24 UTC