Re: Multiple Semantic Web Languages (was: Re: reifying variables)

From: Ian Horrocks (horrocks@cs.man.ac.uk)
Date: 02/12/04

  • Next message: Mike Dean: "Joint Committee telecon tomorrow 17 February"
    On February 12, Sandro Hawke writes:
    > 
    > > I argued long and loud in the W3C WebOnt working group about problems that
    > > using the RDF syntax caused.  This argument didn't go anywhere, so I gave
    > > in and created a partial solution for OWL. 
    > 
    > Do you remember why the WG disagreed with you?
    
    Like Peter, I have argued long and hard about the problems with the
    RDF syntax and have written numerous papers (many jointly with Peter)
    setting out these arguments. I have seen/heard no coherent reply -
    certainly no documentary evidence - only statements of religious
    belief. Of course the problem with opposing religious belief is that
    reasoned argument is useless. In the end, I gave up (I have better
    things to do with my life) and chose to ignore the problem, i.e., I
    accepted the so called layering compromise.
    
    > 
    > > > I've
    > > > sometimes campaigned unsuccessfully for a standard meta-language, in
    > > > which the syntax and semantics of practical data languages are
    > > > described.
    > > 
    > > XML might be the syntax half of this.  Isn't model theory (or even
    > > mathematics) the semantics half?  :-)
    > 
    > Ah, I meant a machine-usable meta-language.  Like java byte code,
    > XSLT, or prolog DCGs -- something you can use to define a new language
    > such that computers can simply use the new language without human
    > assistance.
    
    I have often heard this argument, but I'm afraid that it just doesn't
    stand up. Such a design is an ambitious and probably impossible dream
    (I think you will find that some clever logician/philosopher types
    have proved that it is impossible). RDF certainly comes nowhere close
    to realising this dream - the RDF specification of OWL captures only a
    few trivial details of the language, most of which could be captured
    (much more easily) in an XML schema. To really understand/use the
    language you have to understand the semantics, which is specified by
    the model theory (or rather one of the model theories), not in terms
    of RDF. This is equally true for RDF/RDFS itself.
    
    Talking about cost (which you do below), probably the biggest cost in
    building OWL tools is dealing with the incredibly cumbersome RDF
    syntax - we have a whole document devoted to explaining how to go
    about parsing an OWL/RDF file. Even then, the results are far from
    satisfactory - problems include the fact that streaming is inherently
    difficult/impossible, so there are limits on the size of document that
    can be parsed, and that document structure cannot be maintained, so
    that round tripping a document may cause it to become syntactically
    scrambled and so incomprehensible to a human being.
    
    Ian
    
    > 
    > > > In any case, using an "RDF-Like" syntax is using counterfeit money and
    > > > I suspect is only accepted as the price of admission by mistake.  I
    > > > apologize for not noticing and commenting on SWRL doing this back in
    > > > November.
    > > 
    > > There have been several proposals to make this sort of solution the
    > > official one, including things like dark triples.
    > 
    > "Dark triples" was proposed during a time when RDF Semantics were
    > rather up in the air, I think.    It's less in-order now.
    > 
    > > I would very much welcome a statement from W3C that Semantic
    > > Web languages do not have to be written in triples (or quads, or quints,
    > > ..., as in several recent proposals).
    > 
    > The consortium doesn't have many ways to make statements like that.
    > 
    > Maybe the closest would be in the charter of a new working group or
    > activity.   In fact, the SWRL WG charter I floated a few months ago
    > had some words about the extent to which the language needed to be
    > based on RDF.
    > 
    > What I had in mind were rules like in n3, of the form
    >      forall u1, u2, ...
    >         { rdf graph }
    >         =>
    >         exists e1, e2, ...
    >           { rdf graph }      
    > 
    > where the rule itself did NOT have to be conveyed in RDF.  (I want
    > them to viewable as RDF, and I believe it's possible, but we didn't
    > want the WG to be forced to do that.)
    > 
    > I was interested in seeing if the rule language would itself turn out
    > to be a better RDF serialization language.   
    > 
    > Had this been given to the membership, and had the membership approved
    > it, I suppose you would have had much of what you want.
    > 
    > 
    > In more general terms, I think what qualifies or does not qualify as a
    > Semantic Web language is a complex social question.  What TimBL says
    > in speaches probably counts for a lot, what the consortium members say
    > in WG charters counts for a lot, but so does ... everything else.
    > 
    > And cool, useful, deployed, working systems speak very, very loudly.
    > 
    > 
    > > > You would agree there's a cost to each new language, right?  
    > > 
    > > No.  I do not believe that there is a net cost for most new
    > > languages/representations.  New languages/representations are accepted in
    > > part because of problems with existing languages/representations.  If the
    > > new language/representation does not have benefits outweighing its costs,
    > > it is unlikely to be accepted.
    > 
    > You changed "cost" to "net cost".  Big difference.
    > 
    > You're right about net cost; I was thinking about up-front costs or
    > barriers to entry.  Every new language that agents (eg browsers) need
    > to know makes the cost of developing a new agent higher.  But you're
    > right that the market will decide and people will implement at least
    > dozens of languages if the benefit is there.  And plugins help a bit.
    > Part of the role of the W3C is to try to cut down the number of
    > languages -- one scalable vector graphics language instead of four --
    > to keep this up-front cost in check.    This is something software
    > vendors and user communities really want.
    > 
    > There are also costs to users of upgrading their software, and costs
    > to developers of learning to use a new language.   There are great
    > network effects when more and more people use the same language.
    > Bookstores can label the shelf, instead of just having to hide the
    > books on each little language in Miscellaneous/Oversize.  
    > 
    > So I'm all for people competing with RDF/XML in the hopes they come up
    > with something really better, but we should recognize that it's mostly
    > a winner-take-all proposition, and we can't expect to let a thousand
    > Semantic Web languages bloom (except via the meta-language approach,
    > perhaps).
    > 
    > 
    > All of this is perhaps irrelevant to the original "reification of
    > variables" question, since a logic language with its syntax encoded in
    > RDF can hardly be said to be itself RDF.  It is its own language, with
    > its own deployment and cost issues, even if carried on RDF/XML.  All I
    > was saying was that IF you're going to encode the syntax for the
    > language in RDF, for whatever reason, please encode it in real RDF
    > (with real RDF semantics [plus iron]), not in an RDF-like language
    > with insidious differences from the standard.
    > 
    >   -- sandro
    


    This archive was generated by hypermail 2.1.4 : 02/12/04 EST