notes from today's JC telecon on DAML Rules strawman and plans

From: Benjamin Grosof (bgrosof@MIT.EDU)
Date: 07/08/03

  • Next message: Mike Dean: "Joint Committee telecon tomorrow 15 July"
    % notes from JC telecon 7/08/2003
    % by Benjamin Grosof
    
    Discussion of DAML Rules strawman document sent by Benjamin Grosof
    
    participants:
    
    sandro
    benjamin
    ora
    harold
    said
    ian
    stefan
    mike dean
    pat
    
    
    
    would be good to include also:
    
    XML DTD for RuleML V0.8
    
    "essential RuleML" examples -- e.g., that Said sent out to JC
    
    
    
    Sandro:
    
    would be nice to have a paragraph about what's surprising to those
    familiar with OWL, e.g.,
    - LP semantics vs. FOL semantics
    
    
    Sandro:  another section to have:   test cases, e.g., for entailments
    - e.g., cf. OWL, that illustrate a group decision or agreement
    
    Mike:  don't tie those decisions too closely to use cases
    
    Benj:  should it be sprinkled throughout
    
    Mike:  would have separate document for test cases
    
    consensus:  yes, let's have a separate document for test cases and use cases,
    keep examples in the syntax & semantics document to be expository
    
    Sandro:  but do we have a volunteer for an editor of separate document
    
    Mike:  suggest we have one document to start, to keep it simple,
    then split it up later
    
    what's the goal for this document:
    - Sandro:  would be nice to have a W3C Note
    
    Ian:  early documents of DAML+OIL had single document, overview,
    what's it all about, what would you do with it, why do you want it;
    the whole thing can be relatively short, with a few key components:
    1. handwave-y overview
    2. reasonably brief worked examples illustrating the basic/key features of the
    language (in DAML+OIL, the syntax was specified here)
    3. formal semantics with abstract syntax
    4. formal concrete syntax
    
    Mike:  yes, this is in line with what I'm thinking
    
    Benj:  issue:  What kind of expressiveness for V1? E.g.,
    explicit equality (e.g., among URI's)? basic built-ins?
    
    Harold:  explicit equality in undirected fashion is highly inefficient,
    LP community has basically given up on paramodulation and moved towards
    narrowing, which is a kind of non-ground term rewriting.
    
    Pat:  I agree.
    
    Pat:  Can handle equality via canonical forms, i.e., clustering around
    "equality hubs"; choice of what's canonical within any group of equal stuff
    could be different for different users, but it's OK as long as you can
    translate between them.
    
    Harold:  please send any writeup you have about that
    
    Benj:  how if at all would this change our spec of the rules language?
    
    Harold:  we could omit the symmetry axiom
    
    Pat:  we shouldn't tinker with the definition of what equality is, but
    simply permit systems to be incomplete in their inferencing
    
    Benj:  I agree
    
    Ian:  we can view this (directedness) as an implementation issue
    
    Benj:  what should be the form of our spec of equality?
    E.g., should it be the four axioms from the note I sent, which Pat
    attributes originally to Liebniz?
    
    Pat:  no, it should be directly be defined in the semantics as sameness
    in the model
    
    Benj:  is that well worked out in the literature?
    
    Pat:  yes
    
    some discussion about directed equality, with Harold and Pat
    
    Mike:  it would be nice to have a test case for directed equality
    
    %%
    
    Benj:  wrt syntax:  what kind of syntax:  XML, RDF, human-oriented string?
    two or all of those?
    
    Said:  XML
    
    Mike:  human-oriented string; tempted to use this as exchange syntax too
    
    Sandro:  RDF
    
    Pat:  worry about layering issues in putting it over RDF
    
    Ian:  we can deal with serialization issues later
    
    Benj:  most urgent seems human syntax for communication amongst ourselves
    and our document's examples, and XML syntax for being able to run in
    existing tools, then we can do RDF version later
    
    Sandro:  how about using KIF
    
    Pat:  official KIF
    
    Ian and Benj:  we can keep this pretty simple, e.g., Prolog-y
    
    Benj and Mike:  key issues are how to include URI's/Qnames, and
    unordered argument slots; also lists
    
    Harold's syntax (Prolog-like plus URI's/Qnames, and unordered argument slots)
    with : for Qnames, explicit prefixing of variables, say $ for rest variables,
    different symbol for typing besides ":"
    - action item:  Harold and Said:  will do a draft of that and present
    it at the next JC telecon
    
    %%
    
    Sandro:  what format should the document itself have?
    
    consensus:  let's do it in current stable version of XHTML
    
    %%
    
    Benj:  any other issues we need to add to semantics, besides
    Horn LP semantics plus equality?  E.g., basic built-ins or typing?
    
    %%
    
    issue:  typing
    
    Mike:  typing of literals and also of RDF types (and thus XML Schema)
    
    consensus:  let's keep it simple to start
    
    Pat:  getting data typing right is sometimes non-trivial,
    in OWL it took way more work than expected, eventually settled on a quite
    simple approach:  "context-free", i.e., datatype is locally immediately
    associated with a (rdf:)literal, as opposed to being inferrable
    
    Mike:  maybe rule language needs only to address typing of rules not of
    individuals
    
    Mike and Benj:  a simple approach is to treat types as arity-1 predicates,
    i.e., as classes
    
    Pat:  there's been a big debate about this in CommonLogic, e.g.,
    should types instead be able to be checked immediately syntactically,
    modern inferencing engines are somewhere in the middle, have a special
    reasoner for type inferences (e.g., a RDFS subset of OWL),
    but some details of that are tricky,
    thus has become back-burnered in CommonLogic
    
    Benj and Pat:  a lot of practical reasoners do use a simple sort hierarchy,
    e.g., in LP, Cyc, Snark
    - Pat:  issue of overloading, tree-ness
    
    Benj:  let's make it possible for those who do sorted systems to do it,
    but not get into it deeply ourselves in our spec
    
    consensus:  extensibility to sortedness is thus an objective
    
    %%
    
    
    
    ________________________________________________________________________________________________
    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 : 07/08/03 EST