Re: Layer Cake

Ivan Herman wrote:
> -----BEGIN PGP SIGNED MESSAGE-----

> I must say that I am a bit uneasy restarting the discussion on the
> layercake diagram. There has been a looong discussion (before SWEO came
> on board) on the SW Coordination Group on how to do this without
> radically changing everything and I also had a long discussion with Tim
> on that. It reflects a kind of a consensus right now... This type of
> discussion can drag on....

Apologies for not being around much when that discussion was happening.

I think the premise for the prior discussion may need examining: maybe 
radical change *is* needed?

My take is simple: the layercake is not a useful diagram any more. We 
should stop using it. (Sorry TimBL...). I like many of Tim's diagrams, 
but not this one. The slides at http://www.w3.org/Talks/WWW94Tim/ are 
much simpler and a clever and elegant explanation of SemWeb (which is 
why we turned them into a shirt...). I'm also fond of the original WWW 
proposal doc diagram, see 
http://www.w3.org/1999/11/11-WWWProposal/thenandnow for an RDFization, 
js-prolog-isation, and discussion. But layercake just leaves me cold. I 
can get no real information out of it, and have never found people very 
  enthusiastic about it, except folk showing how their specific 
favourite bit of technology is considered really important to the future 
of the Web (eg. in EU funding proposals or project reports).

For non-technologists, it's a baffling pile of acronyms.

For technologists, it conveys little more than a sense of some 
related/nearby technologies. Aboveness, distance and coveredness 
relations in the diagram don't mean anything very clear.

The main purpose of the diagram I think is now in the past: it served to 
get logic and ontology people on board the SemWeb bandwagon. As they 
say, "Mission accomplished".

Looking at the version in http://en.wikipedia.org/wiki/Semantic_Web

XML does not overhang URI, yet XML can be deployed without URIs.
SPARQL appears irrelevant to Trust, contradicting the SPARQL spec,
eg. http://www.w3.org/TR/rdf-sparql-query/#restrictInQuery where named 
graph facilities are shown as clearly relevant to trust in Web data. 
With or without crypto (we can use PGP against RDF data in all syntaxes, 
and write results of PGP into the quadstore, for example). OWL is shown 
completely covering RDFS, covering up our dirty little secret, which is 
   fragmentation between some OWL flavours and OWL-Full idiom RDFS - 
various OWL tools only consume the DL variant. This is a serious problem 
in the SW technology scene. RIF is shown as a separate parallel stack to 
RDFS+OWL, despite the fact that an RDF-oriented RIF flavour could serve 
as a vocabulary description language, alongside RDFS and OWL. And that 
RIF rules should (see recent CG discussion) be embeddable within 
ontologies, syntactically at least, but ideally with some common 
semantic). RIF also appears as a parallel and disconnected enterprise to 
SPARQL, despite their hopefully at least sharing some functions (such as 
those derrived from XML schema/query). Which BTW is another invisible 
relationship, since SPARQL is not above XML in the diagram. RIF, whose 
current syntax seems non-XML, per 
http://www.w3.org/TR/2007/WD-rif-core-20070330/#head-69087e35ed54052cff4c307536e738af51df17f0 
) is shown partially atop XML, while SPARQL is not, despite SPARQL 
providing a pure XML results format, in addition to the 
function/operator connection. In some ways, RIF ideally sits a lot 
lower; if it succeeds, it could become something like RDF 2.0, as a 
universal and highly-expressive, syntax-neutral information exchange 
mechanism. The multi-billion dollar Web Services industry does not 
appear at all (unless considered merely a part of XML) despite SPARQL 
having a Web service interface to RDF data stores, despite W3C having 
seen fit to create an Interest Group on SW Services, and the Web being 
awash with proposals for using Semantics to describe Services. And if 
the layercake is read as a timeline/roadmap, with earlier milestones at 
the front, and the far future in the distance, ... then user-facing 
applications appear to be the last things on our mind, and something 
we'll get to eventually. Some of the chunks in the diagram are W3C or 
other standard technologies, some are vague topical areas or things 
facilitated by such technologies; some are hypothetical. The use of 
colour is mysterious, and seems unrelated to entity type. Maybe I'm 
looking at the wrong version, but why does a "unifying logic" only 
partially cover SPARQL and (especially) RIF. If it unifies them, ... it 
makes them one. Yet in the diagram, "Proof", the only thing that seems 
to use (or at least, be on top of) "unifying logic", also bypasses that 
logic and sits directly on top of (half of) RIF. Proof seems unrelated 
to SPARQL, despite SPARQL-accessed quadstores being the only current way 
of keeping track of provenance information, ie. who-said-what.

I could go on...

Please give this diagram a graceful burial. It served a useful purpose, 
but things have moved on.

cheers,

Dan

Received on Sunday, 1 April 2007 09:28:12 UTC