W3C home > Mailing lists > Public > www-archive@w3.org > July 2002

Some thoughts on L-Base

From: Sandro Hawke <sandro@w3.org>
Date: Thu, 18 Jul 2002 16:39:52 -0400
Message-Id: <200207182039.g6IKdqx18977@wadimousa.hawke.org>
To: rguha@us.ibm.com, phayes@ai.uwf.edu
Cc: www-archive@w3.org

I'm writing in response to your L-Base documents [1] [2], which I
understand are currently in draft form, perhaps on their way to W3C
Note status.

I've been working on precisely specifying the semantics of our 
"log" vocabulary [3], using essentially the same approach, which has
led me to a few thoughts about your document.

1.  Please specify a machine-readable and human-readable syntax

In [1] you specify a syntax with single letter names and subscripts.
We get the idea, but that's obviously not practical for machine
processing (or sending in e-mail!).  In [2] you're using an ascii
syntax, but it's not the one in [1].  What exactly is the syntax?
Where is it defined?  (More to the point: where are the

I understand you considered using KIF.  I'm using the formula syntax
of otter [4].  You could use TPTP's native format [5], although that's
not very pretty.  Otter's syntax is a lot like what you use in [2],
except it uses "all" instead of "forall", and symbols like "&" instead
of "and".  Examples: "all x y exists z (P(x,y,z) | Q(x,z))" and "all x
(P(x) & Q(x) & R(x) -> S(x))".

It's plain FOL-with-equality (the way I use it at least), so you do
need to say rdf(s,p,o) instead of p(s,o), but I consider that a good
thing; it makes it more clear what's in L-Base and what's in RDF.

I also use rdf_type instead of rdf:type, which otter would require be

Anyway, it doesn't matter what syntax, but I think you do need some
unambiguously parsable ascii syntax....  You *could* use a layered RDF
syntax for FOL, like McDermott et al [6] (I have a translator from one
of them to otter, and plan to write a reverse translator shortly), but
that would probably just confuse people at this stage.

2.  I think it's important to have the text of the urirefs and
literals available.  I recognize this is not supported by the current
MT, but I think that's a mistake.  It's information that's available
to applications and will be used; if we put it out of range of the
formal semantics, then we leave it unnecessarily to the informal ones.

My approach is to translate every RDF document to an L-Base WFF using
only (1) existential variables, (2) the constants zero and
emptyString, (3) the functions succ(), unicodeCharacter(), strCons(),
and webDenotation(), and (4) the predicate rdf() as above.
webDenotation() is the function mapping from strings to the things
denoted by using those strings as urirefs.  I've left out
XML-literals, xml:lang, and unicode versioning issues for now.  (There
are of course, other, equivalent basic sets.  Maybe one that didn't
touch numbers would be better.)

I believe this formalization allows RDF layered languages to have
semantics involving the text of URIRefs, and of literals, while not
breaking anything.   It does increase complexity (for humans and
machines), but it's probably worth it.   For my purposes, some
predicates that Tim B-L uses (eg log:racine) rely on it.

Note that this formalization does not let you ask about "the"
identifier for an object, but just about a string which might happen
to be known to have its webDenotation be the object.   And it
certainly doesn't grant access to L-Base identifiers or anything.

In any case, we need some way to convey the information that all RDF
literals are pairwise not-equal.   If you said how to do that, I missed


Those are probably my two strong opinions.  I mentioned in earlier
e-mail that in some cases consistency of axioms may actually be
provable, eg by finding a model.  

In general, I'm pretty happy & optimistic about this approach,
although axiomatizing some common vocabulary terms is very difficult.

And I haven't a clue what to do with log:contents, which relates a URI
with the strings you obtain by fetching it.  I guess I'll be able to
formally define (or at least constrain) it in terms of some
network-operations ontology.

Have you guys been thinking about how these semantics might be
published, and even automatically retreived, from the namespace
address (or elsewhere)?  

I guess my real question is this: I see how this stuff is very useful
for getting logic layering right, but does it come into play at all
with ontologies for buying books over the web, etc?  I can see it
fulfilling the same role as DAML+OIL or OWL there, sort of.  Should

Or is this really just an on-paper read-by-logicians kind of thing for

    -- sandro

[1] http://tap.stanford.edu/sw/swmt.html
[2] http://tap.stanford.edu/sw/rdfsem.html
[3] http://www.w3.org/2000/10/swap/log.n3
[4] http://www-unix.mcs.anl.gov/AR/otter/
[5] http://www.cs.miami.edu/~tptp/
[6] ftp://ftp.cs.yale.edu/pub/mcdermott/papers/McDermottDou02.pdf
Received on Thursday, 18 July 2002 16:42:27 UTC

This archive was generated by hypermail 2.3.1 : Wednesday, 7 January 2015 14:42:08 UTC