Re: FOL versus Rule Languages - A tutorial

> All- I'm running out the door, not back for a few days -- but looking 
> at the recent mail, Dieter's comments, etc - it seems to me there is 
> a convergence on a "compromise" space occuring -- the charter as 
> written could be changed in a few simple ways to, basically, allow 
> the WG to work out some of the details - the compromise space may 
> live somewhere around here:

Jim,
I don't think that "compromise" is the right word. It implies that there is
a clash of political interests, while I was naively thinking that we were
discussing technical issues.

I would use the term "agreement on technical issues," and I still don't see
it happening.


> 1 - FOL v. model -- the charter as written, with a few small wording 
> changes, woudl allow a WG to produce the following
>    i - a FOL-syntax based language for exchange (Call this Rules Full) 
> - this would essentially be the "XML for the rules"
>   ii - a "semantic" specification for that syntax that might be less 
> precise than needed for rule reasoning  (in fact, I can imagine that 
> an operational or axiomatic semantics may be just fine for this) -- 
> this semantics is not intended for computational use, but for clarity 
> of specification

We are talking about an exchange language, right? As I could understand
from Sandro and others, the goal is to have a language L such that "80%" of
the rule languages out there could be mapped into L in a semantically
preserving way.

Then, with luck, if we have a rule set, R, in a real language L1, and we
want to send those to a guy using another real language, L2, we would send
the L2 guy the message map1(R) (which is a formula in L), and the L2 guy
would then unmap it into R' = map2^{-1}(map1(R). map2^{-1} here is a LaTeX
for the reverse mapping of map2.

Now, map1, map2 are supposed to be semantically-preserving, so running R in
L1 should do the same as running R' at L2.


This is the vision and is very nice. Except that this is still a hard
research issue, not clear whether it has a solution, and thus is not ready
for standardization.

Using FOL for the language L is wrong, as was pointed out repeatedly by
several people. The reason why people propose FOL is because FOL seemingly
has all the needed syntax. But this is a mistake. We need a
*semantically-preserving* mapping, not just a syntactic mapping, and this
is where FOL falls through.

I can't even think what "an operational or axiomatic semantics" (which you
proposed) would look like. The former seems to me like a non-starter and
the latter must be second-order (also probably a non-starter).

This is why I mentioned a RuleML-like approach as a practically achievable
alternative. It doesn't provide for full interoperability, but it supports
partial interoperability.

>   iii - a subset of this language (or, more precisely a profile of the 
> langauge - as OWL DL is a profile of OWL Full) that is intended for 
> computational use - I would imagine that any of the main things 
> people have argued for (datalog, Horn, Courteous LP, etc.) could be a 
> possible basis for this, and the WG would need to work that out - let 
> me call this Rules Comp because I can't think of anything better 
> right now.

Now you seem to be swerving away from the "interchange language" vision to
a different topic -- a Web rules language.
I think we better don't mix these two. They seem to me two different domains
to be dealt with by two different groups.

I don't see how LP-style languages can possibly be viewed as profiles of FOL.
Drawing an analogy with OWL-DL/OWL-Full here is a disservice to OWL.

>   iv. a precise semantics (with a single model as Dieter defined it) 
> for this language
> Note that this would be sort of like WebOnt did with OWL Full and OWL 
> DL - but by knowing in advance this was a possibility, the language 
> design would be much easier -- in OWL the realization that we could 
> do both didn't occur until very late in the game - and if we'd 
> realized going in we might have made some decisions differently that 
> would have made things easier in the end.
> 
> 2 - OWL and this language -- the compromise above would need to cause 
> the words on OWL to change a little - the compatibility with OWL that 
> Sandro specifies would work with the Rules Full language, but would 
> need to be described differently with the restricted subset -- I 
> think just specifying that the WG has to clearly specify the mapping 
> from OWL to Rules Comp (and note, that this can be from OWL Full - 
> i.e. some OWL restrictions for DL may not be needed for the rules - 
> cf. nominals are easy in rules languages)

Perhaps our paper on a multi-stack architecture for the Semantic Web can
help clarify the issues. Dieter mentioned it in his message. I am now
including a URL for it:
    Michael Kifer, Jos de Bruijn, Harold Boley, Dieter Fensel: A Realistic 
    Architecture for the Semantic Web,
    to appear in the RuleML Conference Proceedings, Galway, November 2005
    ftp://ftp.cs.sunysb.edu/pub/TechReports/kifer/msa-ruleml05.pdf


> 3 - NAF v. SNAF - I disagree with Dieter and some others that this is 
> not an important issue - I personally think it will be the critical 
> issue as to whether we end up with a rules language or a new rules 
> language that is the basis for Semantic Web applications (for those 
> who know what I mean, think of the difference in cwm with and without 
> log:semantics and quoting) -- but that said, the charter can leave 
> this vague and the WG can decide -- I would do this by removing the 
> "NAF is out of scope" - but I would leave in something about the 
> ability to close worlds (this needs to be rewritten a bit) -- so in 
> essence I would legitimize the WG dealing with this issue, but 
> wouldnt' demand a specific solution

Removing "NAF out of scope" is a good thing not because it is a
"compromise" but because SNAF+"no NAF"+monotonicity in one document is
contradiction in terms.


>   Anyway, if I was putting on my old WebOnt chair hat, I'd be nervous 
> about how large the charter space was in the above (a narrower 
> charter is easier for a chair), but happier because my hands wouldn't 
> be tied and the WG could produce use cases and requirements that 
> would guide all of the above...

I think that the interchange WG should be different from the web rules
language WG. Moreover, within the web rules language there are several
stacks (see the aforesaid paper): A nonmon rules language, OWL-based stack,
SPARQL. These are best handled as separate WGs, as there is enough work for
all.


	--michael  


> oops - this was gonna be a short message ... sorry
> 
> -- 
> Professor James Hendler			  Director
> Joint Institute for Knowledge Discovery	  	  301-405-2696
> UMIACS, Univ of Maryland			  301-314-9734 (Fax)
> College Park, MD 20742	 		  http://www.cs.umd.edu/users/~hendler
> 
>

Received on Saturday, 27 August 2005 21:08:40 UTC