Re: Semantics101, 2 and 3

>  There are at least three kinds of semantics to consider:
>
>    101 --  tagging data and predicates with their real world types, as in
> xml or rdf
>
>    102 -- model theory based semantics that say what *should* be deducible
> from
>              a collection of rules and facts
>
>    103 -- executable English sentences that document what we mean by our
> predicates
>
>  I think we all pretty much understand the 101 part.

Adrian, I understand it if that means types are needed (a la thesis
9, "language engineering", of the Ten Theses paper [TT]), and it
doesn't mean at all it's an easy thing to do... xml and rdf type
systems are two quite different things by now... ;)

>
>  For 102, there is the question of *which* semantics.  The leading
> candidates would appear to be (a) stratified range-restricted datalog
> with negation [1], or (b) the well-founded semantics [2].  Note that (a)
> is two-valued -- true or false -- while (b) is three-valued -- true,
> false or unknown.  There are many successor papers to each of [1] and
> [2].

Saying these two are the leading candidates looks like a bold
assumption... ;) I'd say these are two candidates, as many other
people have their favourites that differ from those two.
Or, well, I would *bet* many other people have their favourites ;)
There's also the problem of where this rule language
is going to sit: if on top of OWL, or aside, which also has
deep implications on the semantics. There's a very nice paper
by some "usual suspects", that will be shortly published (so for now,
sorry, no pointer until that gets accepted or the authors want
to post it earlier), nicely showing the dangers of underestimating
the "on top of OWL" issue.
And some other people have also doubts this different views
might even be reconciled, whereas another view could be possible,
the "diversity approach" (cf. thesis 1 of the Ten Theses paper,
[TT] ).


>  The need for semantics103 can be seen by reading the jena-dev and other
> SW-related lists.  There, one daily sees smart people getting thoroughly
> confused about the meanings of RDF and OWL constructs and reasoning
> processes when translated into English.
>
>  In our work, we have fielded an online system [3] that meets the 103 need
> without getting deep into NL research.

I completely agree: even more, one doesn't need to go as long
as looking into jena-dev, as this issue in the sweb came up much earlier,
cf. Metalog (http://www.w3.org/RDF/Metalog/ ) with the PNL layer
(http://www.w3.org/RDF/Metalog/docs/pnl.html ). And much earlier in
the industry side too: as you know and as you substantiated with
your own implementation, it's common pratice there to be much more careful
of the "people axis" (http://www.w3.org/RDF/Metalog/docs/sw-easy.html)
and provide for some sort of controlled english/pnl interface for the
customers; in fact, the vast majority of commercial rule engines
nowadays offers such a layer.
Not to underestimate also the companion "visual layer", cf. for
instance visXcerpt
(http://www.pms.ifi.lmu.de/rewerse-wgi4/software/visXcerpt ),
which is a nice analogous approach to climb up the People axis,
this time using the visual layer instead of the verbal one.
Both of these twin aspects are summarized into thesis 10
(Visual and Verbal rendering) in the Ten Theses paper [TT].

Relatedly, the european Network of Excellence on web reasoning
REWERSE (www.rewerse.net) is bringing forth these issues with quite
some strength in its overall view of technologies.

> However, the system depends on
> having a firm basis at the Semantics102 level.  Thus, this is more than
> 'syntatic sugar'  [4].  For example, a two-valued closed world
> semantics102 can support, at the semantics103 English level, an open
> world flavored predicate such as "so far as is known at the moment,
> flight 678 is on time".
Yes, absolutely, the axes are not disjoint but dependant on each
other. And, yes, the verbal/visual layer is not at all "second class",
but ought to be first class, for the implications it brings with it
(and ought to be an explicit SHOULD/MUST in a list of requirements).

-M

[TT] Ten Theses on Logic Languages for the Semantic Web
     http://www.w3.org/2004/12/rules-ws/paper/15/

Received on Monday, 13 June 2005 21:14:30 UTC