Four further issues
I appreciate John Thompson's tactful effort to return the discussion to a
specific judgement about a specific computational artifact. However...
For those of us who have been involved in the efforts to arrive at a
reference upper level ontology (the theme of the onto-std and SUO lists),
four more general issues have emerged that deserve further attention:
1. The relation between "upper level", domain-independent ontological
concepts and axiomatized ontologies serviceable in a given domain.
Chris Menzel writes:
"Obviously, a process ontology in which time is explicitly discrete
would not be interoperable with one in which it was continuous." [He goes
on to assert that PSL is not such a system.]
Leaving PSL aside, since the standard determining "obvious"
non-interoperability is the supposed priority of the continuous-discrete
disjunction, the same caveat on interoperability would seem to apply to
any axiomatized ontology in which space or motion or action or whatever is
discretized in the core definitions and postulates that constrain the
system and its extensions.
This is a thought-provoking outcome. If my memory serves me right, the
fundamental character of the continuous-discrete disjunction was the
consensus of several distinguished contributors to these lists (Lehmann,
Menzel, Guarino, Sowa, Simons, Hayes, among others) at the Heidelberg
ontology conference, and I do not recall dissent from it by any
participants. It was one of several such disjunctions held to be
fundamental, or at least highly useful, for an upper level ontology. Does
it then follow that for any of these fundamentals, if an axiomatized
ontology falls on side A of a fundamental divide, it cannot be
*meaningfully* interoperable with one that falls on side B? And if a lack
of meaningful interoperability is indeed the consequence, this, in turn,
poses challenges for the way in which a necessarily loose upper level
relates to tightly constructed, axiomatized artifacts that fall under it
and are asserted to be compliant with it. These are issues that are
overcome by being addressed, by annotating ontologies with appropriately
differentiated specifications. But it would appear that the
specifications must be quasi-theoretical and conceptual (and I shall argue
at the appropriate time, fully semiotic) and not merely lexical or
syntactic/semantic as has been often suggested.
Have we adequately thought out the issue of heritability between an upper
level ontology and axiomatized sub-ontologies, or is the plan merely to
take the sub-ontologies as unified lumps without attempting to penetrate
their interstices?
2. The practical versus the scientific view of time and process
For the concept of a process itself, two categorially distinct approaches
to the notion of an interval have been raised in the Partridge-Menzel
exchange. Partridge rejects timepoints as "natural" or "intuitive" or,
indeed, as logically correlative to the notion of interval itself. This
can be justified by an underlying notion of processes as comprising
intended activities (each activity having an entelechy that is
intelligible to the participants), versus a notion of them as congeries of
occurrents of those same activities.
Thus, to take the example already proposed, the "beginning" and "end" of a
meeting are markers that are intelligible as meeting-markers only by
reference to the intention of the participants to meet, while the precise
times at which the meeting is called to order and adjourned are markers of
mere extent, ("locative indices" as Peirce would say) from which we can
infer nothing about the content of the interval as a meeting.
I do appreciate that PSL permits an *initial* straddle of
intentional-extensional issues (in the sense used above) by leaving the
verbal definition of activities inclusively vague before they are
rigorously discretized and homogenized within the system. But those
concerned with agreeing on specifications for an upper level standard
should not conflate voluntary with involuntary notions of 'process' if
they wish their ultimate work product to be serviceable to decision
support and to the agent technologies of the future.
3. The kinds of theorizing involved
The computational artifacts that are called ontologies may make use of a
variety of mathematical and logical/semiotic techniques, and appropriate
for their own use a variety of philosophical terms and tropes, but they
are not, as Chris Menzel has pointed out, themselves theoretically
motivated in the full sense. They are what has been called praxiological:
their theoretical element is truncated under a criterion of serviceability
to some pre-agreed goal.
Nevertheless, such artifacts can be improved and their limits can be
scoped by logical and theoretical critique. When computational
architecture replays moves that have had a philosophical history, we can
save a good bit of effort if we know where these lead. Moreover, the
proliferation of bottom up ontologies and the need of large organizations
to integrate legacy systems points to a convergence between a
praxiological and a genuinely theoretical approach: by aiming for what is
theoretically best we shall also have what is most serviceable for those
concerned with integrating and multi-purposing disparate ontologies.
4. The problem of two lists
The decision about whether to use the new ieee or old KSL list has been
assumed to be merely clerical, but there is an element in the formation of
the old standards work that is missing in the new. The previous effort,
constituted as an ANSI ad hoc committee under a more general committee
headed by Tony Saris, began by agreeing that there was a crucial
distinction to be drawn for an upper level ontology between a standard and
an exemplification of it. The standard in the case of an upper level
ontology was properly procedural: it would lay out the key topics-- a
quite extensive list of which was developed-- for specification of any
ontology, but would not address content. A second prong of the effort
would propose a linking of consensus upper level terms with other
artifacts in accordance with the standard and make the result freely
available for general use, along with an explanation of the design choices
made and the overall specifications they exemplified. If the "reference
ontology" were well formed it would become, as Wordnet had become in its
sphere, a de facto standard by virtue of shared use. Under Bob Spillers'
non-directive leadership and with an assist from IBM, whose advanced
technology work he headed at the Santa Teresa lab facility, and further
support from the Tschira Foundation in Germany (inspired by John Sowa's
initial work) considerable progress was made on both prongs of this
project, including work on algorithms for semi-automated alignment of
upper levels which had been designed for differing purposes.
The new IEEE effort, in an apparent effort to cut to the chase, skimps the
procedural dimension. To judge by the declaration of purpose, it hopes to
stipulate the syntactics and semantics (but not the pragmatics) of a
reference ontology, which it will call the standard. The narrowing of
focus to a business/engineering context under ieee may indeed permit
certain simplifications-- such as the elimination of natural language and
cross language artifacts like Wordnet and Penman and the Japnese EDS as
candiates for integration into the reference ontology. But the attempt to
stipulate the reference artifact as opposed to the set of specifications
it exemplifies as *the* standard strikes me as a step backward. It leaves
out the pragmatic dimension of specifying not merely the domain (a
semantic issue) but the purposes to which various sub-ontologies may be
put (search, decision support, knowledge discovery nad logical inference,
cross language work). On the syntactic level, it leaves out the
specification of varieties of ontological inheritance beyond those
embodied in the (first order logic?) artifact. And on the administrative
level, it envisages a stipulative approach to a standard that is more
appropriate to other areas. On the other hand, as the effort is open to
all (as some of the previous meetings were not), its development will
ultimately be influenced by the reasoned consensus of the participants.
If there is overwhelming agreement that the main methodological issues are
settled or irrelevant to the project as now envisaged, then the KSL list
could be reserved for "merely academic" discussion while the ieee list
could be the locus for implementation of a single standard. My view is
that it is premature to think that the methodological/procedural prong has
become obsolete.
* * *
I hope people will forgive the length of this and the previous
comment.
Lee
Josiah Lee Auspitz
lee@textwise.com
17 Chapel Street
Somerville, MA 02144
617-628-6228
fax -9441
Please send attachments pasted within text or in ASCII
Plain Text non-proprietary software.
On Thu, 8 Jun 2000, Thompson, John A wrote:
> The main point of dispute over the PSL Core seems to be whether or not
it can be extended to model situations involving relativity (where there
is no absolute time). Sowa is worried (or convinced?) that it cannot, and
the PSL Core designers are perhaps beginning to admit that no extension to
the Core can model relativity, but they consider that unimportant for what
it was designed to do. > > Is this a correct description of where the
argument stands now? > > John A. Thompson > Mathematics & Computing
Technology > Phantom Works > The Boeing Company > >
-++**==--++**==--++**==--++**==--++**==--++**==--++**== > This message was
posted through the Stanford campus mailing list > server. If you wish to
unsubscribe from this mailing list, send the > message body of
"unsubscribe ontology-std" to majordomo@lists.stanford.edu >