TM Stack: I Beg To Differ

Every now and then I see typical stack diagrams depicting the standards landscape in TM and in RDF land.

While the relationship between TMRM and TMDM is mostly represented correctly, that is, that TMDM constructs can be couched in terms provided by TMRM, the rest of the usual diagrams does not reflect my understanding.

http://kill.devc.at/system/files/tm-stack.png

TMQL, for instance, is not only building on TMDM constructs, its whole formal semantics is based on TMRM path expressions. That may not be visible to the TMQL user, but it is there.

And TMCL is purely defined in terms of TMQL expressions. That, in fact, is one of the significant differences relative to the RDF stack. There, OWL semantics is directly based on RDF model semantics. And SPARQL is defined on the RDF model as well.

I find the TM stack definitely leaner and more consistent, but there is this gaping hole I keep phantasizing about: TMOL.

Constraining

At latest from Larsbot's TMCL tutorial it should have become clear that TMCL is a constraint language. That means that it has a destructive relationship with any topic map:

A TMCL schema defines conditions to be satisfied, only then the map is valid against that schema.

In other words, a TMCL constraint tastes like this:

If something is an instance of CAT, then that instance must also have an occurrence legs with the value 4 and another, utterance, with a value meow.

Otherwise this is bad for the relationship with the map.

Deriving

OWL and other ontology languages work actually the other way round:

If something has an occurrence legs with the value 4 and another, utterance, with a value meow, then automatically it will be classified as instance of CAT.

It is derived and the author does not have to provide it manually.

OWL can only do the deriving part, TMCL only the constraining. (This is somewhat an oversimplification.) That is why there cannot be a straightforward transformation between them.

TMOL could cover the deriving part in the TM stack. If it existed ...

AttachmentSize
tm-stack.svg21.58 KB
tm-stack.png28.44 KB
Posted In

TMOL etc

TMOL should definitely exist, but the other bits need to be done first. There's no reason people couldn't do work outside of ISO on something like TMOL, though...

Regarding the OWL/TMCL (or inference/constraint) relationship I think it should be recalled that constraints/validation is an important business requirement in practice. While inferencing is convenient and impressive, validation is very much a practical necessity.

Lars Marius Garshol (not verified) | Sat, 10/25/2008 - 20:42

Re: TMOL etc

While inferencing is convenient and impressive, validation is very much a practical necessity.

Yup. Both should be there.

rho | Sat, 10/25/2008 - 20:44

TMOL or more general support for inference?

It is nice to have ability to define types, but I think that it is even more important to understand how we can integrate inference into TM-based systems. One important step in this direction, I think, is the standardization of the way we record information provenance.

If we have this, we can build TM-based systems which can utilize various inference modules including OWL-like reasoners.

TMOL can be just a syntax that helps to generate set of constraints and inference rules

Other "screaming" practical issues related to inference: representation of negative assertions: x is not a cat; reasoning in time: x has_a_product y @ 2008

Dmitry (not verified) | Sat, 10/25/2008 - 21:47

Re: TMOL or more general support for inference?

important to understand how we can integrate inference into TM-based systems ...

Implementationwise? The most obvious path is to interpret a logically derived map as a virtual map. More concrete, it means that one implements the map API (could be TMAPI or probably something which better fits the TMDM axes) not with an in-memory map structure, but with a reasoner underneath it. I would call it TM::Virtual::DL if the reasoner is in the DL class.

So if TM::Virtual::DL gets a invocation for give me all classes of that topic and that topic has 4 legs and meows, then CAT will be among the result. Whether this works with backward or forward chaining is an internal implementation issue.

rho | Sun, 10/26/2008 - 10:09

Re: TMOL or more general support for inference?

Other "screaming" practical issues related to inference: representation of negative assertions: x is not a cat;

Well, full support of not is tricky and adoption effectively means to say goodbye to the OWA. There are weaker shades of not in DL.

I am not a strong supporter of OWA (for the Semantic Web it makes sense, though). But to abandon it must be a conscious community decision towards a certain market.

reasoning in time: x has_a_product y @ 2008

Of course, but introducing time might have an impact on TMDM, TMQL and possibly TMCL as well.

rho | Sun, 10/26/2008 - 10:20