From: Benjamin Grosof ([email protected])
Date: 05/13/03
Hi folks,
Here are my notes from today's 5/13/03 Joint Committee telecon (inline'd
below and attached too for convenience).
Mike, you may want to incorporate some of this into the minutes.
Please review for correctness and contribute needed edits -- esp. Pat and
Mike and Gerd and Harold wrt the technical points in the discussion about
SCL and URI's.
% notes from Joint Committee telecon 5/13/03
% by Benjamin Grosof
(Said, Benjamin walked through RuleML examples and semantics)
wrt RuleML semantics:
Pat: seems different in treatment: URI's are supposed to be real names,
so Herbrand doesn't seem sufficient
Benj: OK, let's bookmark that for a discussion later, it's related
to treatment of explicit equality
SCL: (presented by Pat)
has a slightly non-standard semantics
relational and functional extensions are made explicit, e.g.,
wrt applying to themselves and so on
syntax uses AND , IMPLIES
follows KIF
row quantifiers, so can vary-adic relations;
very highly restricted;
can write recursive schema;
can define a trinary
has sequence variables
there's a controversial early draft of...: [Pat will send pter to it]
http://www.altheim.com/specs/xcl/1.0/
XML syntax: 2 very competent people Kamil ___ and Murray Altheim;
they are fighting over fine points of the detail of the design;
incl. relative merits of XPath vs. something
- it's very complicated syntax
wrt RDF:
XML syntax will not be very well related to RDF;
that will need, and probably get, a separate effort
probably isn't really doable due to lack of ability to close off
Benj Q:
relative to textbook FOL predicate calculus,
what are the differences in SCL?
Pat A:
overall:
there's a model theory with abstract syntax
define a SCL language by providing concrete syntax
(there will be a few ready-mades)
and defining relationship of concrete syntax to the abstract syntax
language generalizes in a few ways:
1. roles/slots notation
2. KIF style sequence variables, some of nice KIF abilities with lists
3. can use a function symbol in a relation position, and quantify over
relations and functions, with first-order semantics;
can be translated into first-order using a holds predicate,
but "we" don't bother with a holds predicate;
no underlying comprehension principles;
e.g., can write an axiom stating that a predicate is transitive for example;
e.g., you can say r(r(r)), where first r is a predicate,
second r is a function, and third r is an individual.
has conventional first-order semantics
Benj Q: how related to HiLog extensions in Prolog's, that similarly permit
stating a predicate is transitive
Pat A: don't know. there are indexing and performance disadvantages
to it sometimes
Harold: like that feature
Stefan: I think it's same as HiLog
Benj: ***seems like there's an opportunity to use some of the (2.) and (3.)
features into RuleML
Pat A: yes, just has to do with how one represents atoms
Pat:
4. can associate comments to relation and function symbols too
Benj Q: how does the SCL semantics relate to semantics aspects of
philosophical issues on representing URI's?
Pat A: doesn't really relate (yet), virtually nothing in abstract syntax
about this, can allow non-URI's
Pat: hope that one could axiomatize notion of URI's in SCL, but probably
won't have time to do that
Benj Q: how is this formulated in relation to logic? e.g.,
standardized dereferencing notion from modal logic theory is known to
be problematic in presence of multiple interpretations with non-fixed
domains
Pat A: have found that there are
3 types of reactions:
1. e.g., TimBL: say there's a clear notion that needs to be discussed
2. e.g., Roy Fielding, editor of RFC 2396 (right number?),
would like to avoid it;
has a REST model, has a notion of what gets spit out in response to
HTTP GET (e.g., a webcamera), informal notion of representation
3. e.g., Dan Connolly:
oh dear, seems complicated, needs discussion
Pat agrees with (3.), seems very related to social meaning in RDF;
there seems to be notions in people's heads, e.g., TimBL's, or Roy's notion
of "has an identity", that need to be more clearly articulated
Benj Q: why not be shallow, at least for time being,
wrt logical account of URI,
since it's so murky what is at other end
when dereferencing, and just state info about URI's using logic
Pat A: basically agree, but believe it's worth trying to understand
Web account of how you name something, "since Web names do something"
Harold: related to object identity issues in object-oriented,
since content may change
Benj: don't necessarily need a new logic/KR just because we're
on the Web
Pat: yes, and not suggesting we stop all KR til this is figured out.
But this is a well-known problem in KR: known as "symbol grounding".
"The Web works, let's try to figure out why."
Benj: the web works for same reasons why lots of stuff with social
agreement and use patterns works
Harold then Gerd and Mike: W3C'ers talk about URN's vs. other URI's:
some (?URN's, "stable" URI's or "persistent" URL's) are supposed to be
more fixed.
Pat: there's an agreement in principle to include PROCEDURAL ATTACHMENTS,
but has not been designed in much detail yet:
5. can have procedural attachments to relation and function symbols;
at least, some way to associate arbitrary annotating stuff,
e.g., text for comments, that is semantically transparent;
design idea of a wrapper around any well-formed SCL expression,
which can contain a text/other attachment, and the SCL meaning is just the
SCL expression; i.e., arbitrary tagging or annotation.
Benj: you get that wrapper notion for free in XML
Pat: yes, but it's nice to put it into the abstract syntax for non-XML
syntaxes as well
________________________________________________________________________________________________
Prof. Benjamin Grosof
Web Technologies for E-Commerce, Business Policies, E-Contracting, Rules,
XML, Agents, Semantic Web Services
MIT Sloan School of Management, Information Technology group
http://ebusiness.mit.edu/bgrosof or http://www.mit.edu/~bgrosof
This archive was generated by hypermail 2.1.4 : 05/13/03 EST