Re: comments on Editor's Draft of UCR -- motivates links

The "default behaviour" requirement is not at all about default rules.  It
is instead about handling constructs that are not understood.

For example, one such default behaviour would be to turn any non-strict
rules (default rules, negated rules, or whatever) into strict rules for
systems that only have strict rules.  Another default behaviour would be to
just ignore such rules.  

peter



From: Leora Morgenstern <leora@us.ibm.com>
Subject: Re: comments on Editor's Draft of UCR -- motivates links
Date: Wed, 28 Jun 2006 15:48:36 -0400

> 
> Peter, 
> 
> 
> > > > Default behaviour
> > > > 
> > > >    RIF must specify at the appropriate level of detail the default
> > > >    behavior that is expected from a RIF compliant application that
> > > >    does not have the capability to process all or part of the
> rules
> > > >    described in a RIF document, or it must provide a way to
> specify
> > > >    such default behavior.
> > > > 
> > > >    not motivated by Use Case 2.2, Use Case 2.4, Use Case 
> > > > 2.5, Use Case
> > > >    2.6 
> > > 
> > > From use case 2.6:
> > > 
> > > "This use case illustrates how the RIF makes it possible to merge
> > > rulesets from diverse sources in diverse formats into one rule-based
> > > system, thereby enabling inferences that might otherwise have
> remained
> > > implicit."
> > > 
> > > This rule-based system may get rules that it can't (completely)
> process
> > > and involves important medical decisions so I'd say that default
> > > behavior is motivated.
> > 
> > I still don't see this as part of the use case.  I don't see how the
> use
> > case speaks to partial understanding of rule sets.  On the contrary, I
> > would say that this use case speaks to the necessity of *complete*
> > processing of rule sets, because otherwise some important rule might
> no be
> > processed accurately and someone might die.
> > 
> 
> I believe the point David is making --- and with which I agree --- is
> that many medical rules are better expressed as default rules than as
> universal rules. For example, in this use case (2.6), the rule 
> 
> "If an oral monotherapy at recommended doses of a sulfonylurea or
> biguanide, combined with lifestyle changes, is ineffective, then the
> monotherapy should be replaced by an oral bitherapy." 
> 
> would be better expressed as a default: 
> 
> "If an oral monotherapy at recommended doses of a sulfonylurea or
> biguanide, combined with lifestyle changes, is ineffective, then the
> monotherapy should *usually* be replaced by an oral bitherapy." 
> 
> using whatever means you prefer to express defaults (eg., abnormality
> predicates).  There are, after all, probably exceptions to this rule:
> there many be another class of drugs worth trying as a monotherapy, or
> the patient's particular medical situation may contradict an oral
> bitherapy and may necessitate insulin injections. And one would want to
> be able to specify these exceptions in the rules. 
> 
> I agree that slightly modifying this use case would better highlight the
> need for defaults, and I'd be happy to make such a modification for the
> next draft. 
> 
> By the way, use case 2.10 shows a clear need for defaults. That case
> includes the rule:   
> 
> "Every movie produced before 1930 is black and white." 
> 
> Actually, *most* movies were black and white, but some were in color;
> there were different coloring technologies introduced early on,
> including an early version of Technicolor introduced in 1917. Not that
> I'm a movie buff or anything ;) ; the point just is the need for
> defaults. 
> 
> 
> > > > OWL data
> > > > 
> > > >    RIF must cover OWL knowledge bases as data where compatible
> with
> > > >    Phase 1 semantics.
> > > > 
> > > >    not motivated by Use Case 2.4, Use Case 2.6
> > > 
> > > While 2.6 doesn't specifically mention OWL, it does mention
> ontologies
> > > as a possible data source. I think parts of SNOMED have been
> translated
> > > to OWL, but I'm no expert.
> > 
> > Sure, but you can write (very, very, very) simple ontologies in just
> RDFS,
> > so there is no demonstrated need for OWL. 
> 
> I believe the point is that  parts of SNOMED have actually been
> translated to OWL, or that there are intentions to do so. If parts of
> SNOMED are or will be in OWL, then the RIF will have to handle OWL in
> order to use SNOMED. Whether or not these ontologies could have been
> represented in a simpler language is irrelevant. 
> 
> 
> > 
> > Note that many of my complaints could be addressed by appropriate
> > modification of the use cases. 
> 
> Absolutely.
> 
> Leora

Received on Wednesday, 28 June 2006 21:15:52 UTC