Re: [Sem] Syntax and model-theoretic semantics: a complete proposal

> On 12. Apr 2023, at 10:18, Franconi Enrico <franconi@inf.unibz.it> wrote:
> 
> I extended my original proposal to include a fully opaque case, so that it can represent both the fully opaque case (syntactic predication) and the fully transparent case (semantic and modal predications).  
> Now the proposal comes with a completely specified abstract semantics, etc.
> This proposal comes in three flavours, depending on the syntax chosen: 
>     1. the Community Group syntax with TEP as it is described in its final report (wiki: CG syntax specification), 
>     2. the variant which distinguishes syntactic from semantic quoted triples (wiki: alt syntax specification), 
>     3. the variant with a quoting operator which gives a syntactic reading to arbitrary terms (wiki: alt syntax with quoted terms specification).
> There is a wiki page describing each variant of the proposal.
> —e.

Not commenting on details of the formalization, but from what I understand I like variant 3 the most. I wonder why you still keep the very shaky concept of Modal predication around, but that’s another issue.


What I find really interesting is that you offer a possibility to define referential opacity only on specific terms within a triple term. Would it be possible to go one step further and introduce the notion of a "syntactic term" into RDF proper, meaning that not only triple terms but also IRIs in stated triples can be defined to be referentially opaque? E.g. 

    :LoisLane :loves ':Superman'. 

would then allow to describe that Lois Lane loves Superman, but not Clark Kent as she is not aware that the two are the same person. Even adding that knowledge to the graph explicitly

    :Superman owl:sameAs :ClarkKent .

would not entail that 

    :LoisLane :loves :ClarkKent .  # NOT entailed

Annotating those statements would faithfully capture these subtleties:

    << :LoisLane :loves ':Superman' >> a :ComicPlot .
    << :Superman owl:sameAs :ClarkKent >> a :BackgroundKnowledge .

A different use case, e.g. syntactically strict provenance capturing, may still go for full referential opacity of the triple term (including its eventual use of syntactic terms):

   '<< :LoisLane :loves ':Superman' >>' a :CodeExample .  
                           # note the ':Superman' syntactic term
                           # nested in the syntactic triple term

How to handle blank nodes in referentially opaque triple terms is another question. As I argued elsewhere I’d rather keep them opaque as there meaning is defined not only by the triple term at question but by all triples they occur in. But this is an orthogonal discussion.


This approach would offer two important feature:
- it separates two orthogonal issues that are each complex enough on their own: statement annotation and referentially opaque references. Separation of concerns is always a desirable property of good system engineering and here it wouldn’t add much complexity. To the contrary it would make the IMO quite ill-conceived TEP mechanism obsolete.
- it makes the concept of referentially opaque references available in stated triples, not only in quotations of such triples. This would really expand the expresivity of RDF, and in an intuitive way that doesn’t make anybody wonder what’s going on, i.e. why expected entailments suddenly don’t hold anymore.


Best,
Thomas

Received on Wednesday, 12 April 2023 21:18:42 UTC