Re: Which semantics?

> On Feb 15, 2017, at 5:43 AM, Sebastian Samaruga <> wrote:
> OK. But sorry again for my lack of knowledge but does this mean that 'semantic' inference of the kind of 'inferring' that:
> <>
> (is the same as)
> <>
> is not possible without resorting in previous knowledge

If I understand you, indeed it is not possible without previous knowledge; but not of dictionaries, but rather of identities. Quite a lot of the information stored in RDF is exactly of this form, expressed using the relational URI owl:sameAs, in the form of an RDF triple.

> or dictionaries or, even worst, NLP over those URIs? Not even to mention 'inferring' identity between 'The capital of France' and 'Paris' or 100cm / 1meter.

All such identities must be expressed directly or indirectly in some ontology, probably in the form of an RDF graph. For the first example it might be more useful to simply say that Paris (Subject) is the capital city of (Relation) France (Object) in a single triple, rather than use owl:sameAs. Identity of measure terms has been tackled by ontologies of units and measurements, such as OM (

> Another kind of inference that simply concatenating datasets just not solve is that of 'ordering':
> Joe takes his car out.
> Joe washes his car.
> Joe takes his car in.
> How if the statements comes in any order one could reason about the correct sequence. This will be indispensable for propositional like logic and inference.

Actually not (indispensable, that is). Logics typically do not use the ordering of statements to encode content. (There is a good reason for this, having to do with how logical inference rules can be stated, but it would take too long to explain.)  To express meaningful orderings, especially time orderings (as here), one must describe the ordering relations explicitly. There is a huge literature on how to use logics to express such things, far more than I can summarize here, but one way to proceed would be to describe things one might call ‘events’ (takings in and out, washings, etc.), classify them into types or categories (corresponding roughly to the English verb) and associate them using ontological relations to times (to get the orderings), subjects, objects, and possibly such things as locations (where it happened) and reasons (why it happened). All of this can be directly written as RDF graphs and reasoned about using RDF or OWL reasoning engines. 

All of this has been thought about very hard by a very large number of people for about 60 years now. You don’t want to try to re-invent it all, would be my advice. 

Best wishes

Pat Hayes

> Best,
> Sebastián.
> On Feb 14, 2017 4:20 PM, "Martynas Jusevičius" < <>> wrote:
> Sebastian,
> I think it is useful to think about the merge operation between datasets.
> Here I mean a "physical" merge, where records with the same
> identifiers become augmented with more data, when multiple datasets
> are merged together. A "logical", or "semantic" merge, with vocabulary
> mappings etc., comes on top of that.
> So if you take the relational or XML models, there is no generic way
> to do that. With RDF, there is: you simply concatenate the datasets,
> because they have a stable structure (triples) and built-in global
> identifiers (URIs).
> That said, you should try approaching things from another end: start
> building a small but concrete solution and solve problems one by one,
> instead of overthinking/reinventing the top-down architecture. Until
> you do that, you will probably not get relevant advice on these
> mailing lists.
> On Tue, Feb 14, 2017 at 6:21 PM, Sebastian Samaruga < <>> wrote:
> > Sorry for me being so ignorant. But what could be called 'semantic' (in the
> > sense of 'meaning', I suppose) for the current frameworks, at least the
> > couple I know, available for ontologies of some kind if they could assert
> > between their instances which statements and resources are equivalent (being
> > them in a different language/encoding or different 'contextual' terms for
> > the same subjects for example).
> >
> > Another important lack of 'semantics' is ordering (temporal or whatsoever)
> > where a statement or resource should be treated at least in relation to
> > their previous or following elements.
> >
> > If my last posts where so blurry is because I try to address some of this
> > issues, besides others, trying no to fall in the promise that adhering to
> > one format will free us all of any interoperability hassles. Remember a
> > similar promise from XML: "All we have to do is share DTDs and
> > interoperate". I'll still trying to give the format a twist (RDF Quads) but
> > I'll publish a Google Document open for comments.
> >
> > Best,
> > Sebastián.
> >

IHMC                                     (850)434 8903 home
40 South Alcaniz St.            (850)202 4416   office
Pensacola                            (850)202 4440   fax
FL 32502                              (850)291 0667   mobile (preferred)

Received on Saturday, 18 February 2017 06:01:39 UTC