[sc34wg3] SAM-issue term-scope-def

[Bernard Vatant]

> * Bernard Vatant
> |
> | I understand what you say here as:
> | "A is valid *if and only if* scope applies" ...

*Lars Marius Garshol
> Yes. Valid, not necessarily true.

I'm lost here. Could you explain what practically will make the difference between valid
and true for an application?

> If the topic map has "X instanceOf Y
> in scope Z", and you set context scope to "W", and ask "is X an
> instanceOf Y" the system will have to say "no".

Hmmm. I thought scope on "instanceOf" was not allowed, something I did not understand well
BTW, we had a debate on that and you explained me why. Has it changed since?

Anyway let's assume it. I maintain in this case that the only logical sustainable answer
of the system there should be: "I don't know". Say for example "X is an instanceOf expert"
in the scope "Markup Languages". The scope type there being the field of expertise. Now I
switch the scope to "Astrophysics". To the question "is X an expert in the scope
Astrophysics", I have no elements of answer, and there is no reason to answer "no" more
than "yes".

> | That does not seem consistent with what you say elsewhere; I
> | understood you supported:
> |
> | "If scope applies, then A is valid"
>
> That was when we were talking about truth. But I don't think truth and
> validity are the same.

That's too subtle for me. In my primitive mathematical view of the world, truth is a value
(generally 0 or 1) assigned to an *assertion*, whereas validity is conformance of an
*inference* to a set of axioms and rules.
The way I see assignment of scope in terms of truth and validity is for example as
following:

-- An assertion A can be attached a truth value v(A) in {0,1} (forbid for the moment other
values not to get stuck in fuzzy logic)
-- Assignment of scope has also a truth value v(S) in  {0,1}

Valid inferences about v(A) and v(S) can be

1: If v(S) = 1, then v(A) = 1
2: If v(S) = 0, then v(A) = 0

We have to agree on which of those two are valid. Only 1? Only 2? or both?

Or maybe you give another meaning to truth and validity, that is not valid/true in my
scope :o)

> There is a little bit of natural language set theory and logic in
> there now, but I've chosen to use natural language for those things
> precisely for this reason. But in a sense you are right. Some things
> are needlessly much harder to say without maths.

As I used to say, maths were invented precisely to that - help say things in a clear and
accurate manner.

> | e.g. if scope declaration supports a necessary condition, or a
> | sufficient condition, or both ...
>
> Have you read the document?

Sure. Maybe I've not the latest version (mine is dated 2002 May 06) because I have not
been able to connect to any address under
http://www.y12.doe.gov/sgml/sc34/ for quite a while.
Mary Nishikawa reported the same trouble in Montréal, and it seems that somehow this
server refuses some types of connections.

> If not, please do, and give concrete
> comments on it. If you have, how would you work something like this
> into it? That is, could you turn that sentence into an instruction
> that an editor could have some hope of following?

OK. I'll try to do that.

> | but what if both syntax and "natural language specification divorced
> | from maths" contain such hidden inconsistencies that nobody will
> | never be able to build a consistent model of them? So even if SAM is
> | not expressed in maths language, which I fully agree with, it has to
> | be checked for consistency of its implicit underlying model rules.
>
> How? By whom? What sorts of inconsistencies could there be? What would
> their effects be?

Absent formal model, inconsistencies unfortunately will not appear until you begin to use
rules to build inferences, using TMQL or TMCL, or through any algorithm trying to drive
inferences from the assertions expressed in a topic map. Effects could be e.g. that the
same or different applications using the same topic map drive contradictory conclusions
from it, like : "This user must see that occurrence" and "This user must not see that
occurrence", or "This topic belongs to that class" and "This topic does not belong to that
class". Basically, the effect of an inconsistency is that you can infer from the model two
contradictory conclusions.

And my concern is that is the kind of things we are at the risk to find with the current
fuzzyness on scope, or the absence of templates (another thread).

Bernard