Triple terms in subject position

Dear all,

On our last meeting the people in favor of triple terms in subject position got the „homework“ to explain their reasons.

I try to do this here (and I try to keep it short).



I’d like to have symmetry for reasoning: 
I think that in all RDF related formats, the consequences of reasoning should be expressible. So, I would like that

:isReifiedBy  owl:inverseProperty rdf:reifies. 
_:x rdf:refies <<( :s :p :o)>>.

Entail (in some rdfs entailment enhanced with support for inverse property) that
<<( :s :p :o)>> :isReifiedBy _:x.

As this aspect is relevant for RDF/S semantics as well, we already agreed on the compromise that the semantics is defined on „symmetric RDF“.


I have problems with the counter argument for graph terms in subject position:
I know that we do not allow literals in subject position because we would want to avoid that people write statements like

„01“^^xsd:intger  :smallerThan 2.

or

„Bob“ foaf:knows „Sue“.

Mainly, because „01“^^xsd:intger refers to the integer 1 and  „Bob“ refers to the value of the string (applying lexical to value mapping). Another reason is that allowing literals in subject position  could motivate users to not use IRIs (my:Bob instead of „Bob“ to get the right Bob here). (I also admit, that I think that this is not on us to restrict users, but that is a different discussion).

The argumentation for not having triple terms in subject position as I understand it is quite similar.

We say that

<<(:s :p :o)>> :saidBy :bob.

is problematic because <<(:s :p :o)>> refers to the statement itself and not to any further meaning of it. But in my opinion, this „statement as such“ notation is still very weak in our semantics (despite the infectivity, which I also dislike, but that is another topic) and as long as that is the case, we do not really have a problem here. Furthermore, while I would expect that many users would make general statements about the same literals if they were allowed to  (1 is after all common number), I do not expect the same for triple terms.  All components of triple terms are still represented by IRIs. I therefore think that it is not that likely that statements with the exact same triple terms <<(:s :p :o)>> in subject position will be widely used throughout the Web (I admit, that for example axiomatic triples like <<( rd:type rd:type rdf:Property)>> might be an exception here).  So, to me, this restriction looks and feels rather artificial. 

As I said in our meeting, I am willing to compromise here, but I would personally prefer to not put restrictions where they are not absolutely necessary. 

I am pretty sure to get many counter arguments here. Should we use the GitHub issue (I recommend https://github.com/w3c/rdf-concepts/issues/138 and I can post my reasons there as well) to further discuss or better continue on this list (I am afraid of spamming people on the list with e-mails)?

Kind regards,
Dörthe