[sc34wg3] Subjects, role players, and user-defined association types

[Steve Pepper]

SRN wrote:
>    If you're still with me, here, you're ready for the
>     punch line: XTM *also* allows topic map authors to
>     specify *arbitrary* kinds of relationships, by
>     means of <association>s.

I agree, but I think <occurrence>s and <baseName>s should be
included here, for completeness, along with <association>s
(since we are talking in terms of the SAM). Even though they
allow progressively less arbitrary kinds of relationships,
they still provide the user with the freedom to define their
own "subject-conferring" semantics.

>   This is where the RM
>     comes into 13250 (i.e., into HyTM and XTM).

You make it sound as if this is *the* (one and only?) purpose
of the RM with respect to TM models such as the SAM. Is that
really the case?

>    What if some of those user-defined relationship
>     types are supposed to confer subjects on some of
>     their role players?  How do we tell when such
>     subjects are the same, and therefore must be
>     merged?  Neither 13250 nor the current SAM faces up
>     to the possibility:
>
>     * that a user could define an association type
>       whose instances determine the subjects of one or
>       more their role players, and
>
>     * that more than one topic may thus have conferred
>       upon it the same subject, and
>
>     * that therefore such topics need to be merged.

I wonder what you mean by "faces up to the possibility"?

Neither 13250 nor the current SAM *preclude* that possibility.
(I know, because I exploit it regularly, for example when I
say to my application: "If topic A and topic B have identical
occurrences of type 'email address', they must be merged".)

What neither 13250 nor the current SAM do is to provide a
*standard means* for expressing this semantic such that it
can be understood across applications.

I agree this capability is something we need, but do we need
the RM to get it? Didn't we already get very close to
providing it in Baltimore - using a simple PSI, and without
resort to the RM in any way at all? I am, of course, referring
to the TNC discussion.

The resolution of the TNC issue basically allows a "user [to]
define an association type whose instances determine the
subjects of one ... [of] their role players". (The phrase
quoted is from the first of SRN's bullet points, above.)

The solution, for those who have forgotten, is to define
a PSI whose semantic is such that any base name (of a type
that is an instance of that published subject) "determines
the subject" of the corresponding topic (in the sense that
identical values triggers merging).

Now, we have not yet defined that PSI, nor have we said
what happens if someone creates an occurrence type or
association type (rather than a base name type) that is an
instance of it. If we solve this in the most general possible
way, what we essentially end up with is DAML's "unambiguous
property" :

    if P is an UnambiguousProperty, then if P(x, y) and
    P(z, y) then x=z.
    aka injective.
    e.g. if firstBorne(m, Susan) and firstBorne(n, Susan)
    then m and n are the same.

    (http://www.daml.org/2001/03/daml+oil)

We could even give it the published subject identifier

    http://www.daml.org/2001/03/daml+oil#UnambigousProperty

and be done (although we wouldn't then conform to the OASIS
PubSubj TC's recommendations).

The question all this leads up to is, why can't we simply
do all this through TMCL, which we know we need anyway in
order to express other kinds of constaining semantics than
those that relate to subject identity?

Steve

P.S. To avoid misunderstandings: The hidden text behind
this message (if there is one), is not "we don't need the
RM", but rather "please help me understand exactly how the
RM will enable information aggregation across TM Models".

--
Steve Pepper, Chief Executive Officer <pepper@ontopia.net>
Convenor, ISO/IEC JTC1/SC34/WG3  Editor, XTM (XML Topic Maps)
Ontopia AS, Waldemar Thranes gt. 98, N-0175 Oslo, Norway.
http://www.ontopia.net/ phone: +47-23233080 GSM: +47-90827246