Re: soundness and semantics

François Bry wrote:

>It seems to me that behind the "soundness" discussions, the "semantics"
>issue is re-emerging.  My understandsing is as follows:
>
>1. Interchange of rules and rule sets originally specified after
>different semantics is needed. Therefore, the possibility of RIF rules
>and rule sets with different semantics should be a requirement.
>
>2. Such an interchange is only possible if the original semantics of the
>rules and rule sets to be interchanged is specified. Therefore, the
>possibility to specify a semantics for a rule or rule set should be a
>requirement.

Although there has been email discussion subsequent to this, I think
the above text nicely states minimal RIF features required in
order to support interoperability and rule type/language scope 
(coverage) goals that we currently have for RIF.  These requirements
seem obvious to many, but can we please explicitly take a decision on
this and record it.  It is not necessary to refine what we mean by
"more than one semantics" (2, 3, or greater than 5) at this point.
Nor do we have to say *how* these semantics can be discriminated in a
set of RIF syntax, just that there must be some way to do so.  These
other issues are refinements that will come later.  Let's nail down 
what we can now, to avoid revisiting this over and over again.

In Franks diagram, 1 could be a refinement of Formal Semantics supporting
the Coverage Critical Success Factor.  2 is already in his diagram
as Markup of Semantics.

>3. In most cases, the specification of a semantics for interchange
>purposes is only possible or meanningfull if this specification is
>simple enough (ie expressible in a few words, not in ten thousand of
>lines of code) and abstract enough ("abstract" being understood here as
>"factoring out some aspects").
>As a consequence, simple and abstract specifications of semantics for
>RIF rules and rule sets should be a requirement.

This last bit describes a qualitative aspect of a RIF feature, rather than
a feature itself.  I am not sure how one can compare abstractness, much
less understand what "simple enough" or "abstract enough" is.  Thus while
this seems a good goal, I don't see it as a requirement.  

-Evan

Evan K. Wallace
Manufacturing Systems Integration Division
NIST

Received on Wednesday, 17 May 2006 17:21:27 UTC