ISO 11179; Basic Semantic Repository; Z39.50/STAS

Fritz Lehmann <flehmann@orion.oac.uci.edu>
Date: Tue, 25 Jul 1995 20:21:08 -0700
From: Fritz Lehmann <flehmann@orion.oac.uci.edu>
Message-id: <199507260321.AA28682@orion.oac.uci.edu>
To: bsr!fritz@rodin.wustl.edu, bsr@utu.premenos.com
Subject: ISO 11179; Basic Semantic Repository; Z39.50/STAS
Cc: bsr@premenos.com, cg@cs.umn.edu, edi-new@tegsun.harvard.edu,
        rhw@sprint.bellcore.com, srkb@cs.umbc.edu,
        x3l8-news@bcr.cc.bellcore.com, z3950iw@nervm.nerdc.ufl.edu
Sender: owner-srkb@cs.umbc.edu
Precedence: bulk

     Paul Hawes wrote:

>You mentioned the "content" chapter of 11179. I do not
>know to which of the six parts you are referring.

     I was referring to the Taxonomy Annex in my draft.  This contains
a tree-structured taxonomy of upper and real-world "ontological"
concepts.  I have some critical suggestions for the taxonomy, mainly
based on the unsound effort to force the taxonomy into a tree structure;
all of my studies of semantic networks, taxonomies, subsumption hierarchies,
determinables, faceted classifications, thesauri, etc. indicate that a tree
structure is too restrictive, and causes some very arbitrary subdividing
decisions.  A direct-product of trees is better, but still not good
enough.  What is needed is a general ordered set (partial order or poset),
which may still have independent "factors" (facets/dimensions).

     I was disturbed to hear from someone that you might choose no "content" at
all in 11179.  Although people may never agree on the "one true ontology", you
still need to have some basic "semantic primitives" that people agree on,
in order to use these to build up the meanings of desired data elements. 
I don't mean any ultimate philosophical primitives -- just simple concept
definitions that people can agree on in advance.  These should be very
careful constraints on meaning although they don't have to specify the
_precise_ limits of a semantic class (usually impossible anyway outside math).

    If 11179 fails to have any standardized basic "semantic content" at all, 
who will provide it?  No _automated_ integration can occur without machine-
usable semantic specifications of data elements and value-codes; this
more ambitious requirement depends on there being a pre-existing, agreed-
upon definition of some basic conceptual building-blocks, even if only
in natural-language definitions.

     Thanks for clarifying the relation of ISO 11179 to the EDI BSR
(Basic Semantic Repository).  Denis Hill, head of the UN BSR committee,
has said things a number of times in email that seem to warn against
excessive attention to correctness and detail, so that the BSR effort can
move forward fast as a practical matter.  This is quite legitimate, but
I am reminded that the same "practical orientation" (dispensing with
careful analysis of meanings and coordinating of different sources)
is what led to the X12 and EDIFACT mess in the first place, and it has
also resulted in the somewhat amusing uncoordination within the
STAS Scientific and Technology "attribute set" of the Z39.50 standard.
(STAS was offered by Terry Winograd at SIGIR-95 as an example of
how _not_ to combine data elements).  The path between "too-academic"
and "too-ad-hoc" is a difficult one; it's important to be practical and
fast, but there's nothing practical about finding later that you did 
things fundamentally wrong.  If things like ISO 11179 are deemed too fancy
and "impractical" for use by the BSR effort, then I'd get worried.

                          Yours truly,   Fritz Lehmann
GRANDAI Software, 4282 Sandburg Way, Irvine, CA 92715, U.S.A.
Tel:(714)-856-0671               email: fritz@rodin.wustl.edu
=============================================================