[topicmapmail] Re: How to best model relations with mode (assertive, modal, intention, negation) with topic maps?

[Murray Altheim]

Quoting Alexander Sigel <sigel at wim.uni-koeln.de>:
>
> Dear Jack (and dear topic mappers):
>
> you had kindly replied with some remarks sent 2006-10-23 to my   
> question on how to best model some assertions with topic maps.
>
> Here is my reply [AS]:
>
> [AS]
>>> E.g. "cancer cells IS_IN_PLACE_OF [2] lungs"
>>> means that lungs MAY INCLUDE cancer cells, whereas "cancer cells
>>> IS_IN_PLACE_OF [1] lungs"
>>> would mean that all lungs include cancer cells.
>
> 1. PURPOSE
> -----------
> [JP]
> What is the purpose or intention or goal of a topic map containing   
> assertions of the type "cancer cells IS_IN_PLACE_OF[2]lungs"?
> Would this be a "tutorial" topic map, a "diagnostic" topic map, an   
> index of information resources about cancer?
>
> [AS]
> The right person to ask about the rationale or usage scenario of   
> such an ICLO-based system is Prof. Dr. Winfried Schmitz-Esser [SE],
> the originator of this idea.
[...]

[I've rather resisted commenting on this thread previously,
but here's my 2c, for what it's worth.]

Actually, the right person to ask in this instance is usually
not the "knowledge expert". Prof. Dr. Schmitz-Esser may be
unique in having expertise within the specific domain (in
this case, biomedicine) a functional, pragmatic understanding
of the epistemological underpinnings of the explicit modeling
paradigm being used, as well as how this is mapped into the
specifics of the concrete representation (e.g., a Topic Map),
but my guess is that this is not the case -- I don't rightly
know. It's a lot to ask of one person.

The person who is likely the most qualified to comment on
how the Topic Map accords with reality is the person (the
"knowledge engineer" or "ontological engineer", for lack of
a better title) who designed the model, i.e., the person who
acted as the intermediary between the expert and the
implemented modeling system.

One of the reasons I grew (rather quickly) frustrated with
the field of "Knowledge Representation" is that it is very
rare to find anyone in the field with any background whatso-
ever in epistemology; most are some kind of engineer, even
if they don't realize it or won't admit it. Even most of
the logicians, apart from John Sowa and a few others, seem
relatively ignorant of the limitations of the systems they
develop, though to John's great credit he has gone to great
effort to educate people in this regard. Some in the field
of computational linguistics -- e.g., Geoff Nunberg at
Stanford comes to mind -- certainly appreciate the enormous
difficulties. [Some of you have likely heard this rant from
me before.]

As with Jack, when I see something like:

>> E.g. "cancer cells IS_IN_PLACE_OF [2] lungs" means that lungs MAY  
>> INCLUDE cancer cells, whereas "cancer cells
>> IS_IN_PLACE_OF [1] lungs" would mean that all lungs
>> include cancer cells.

my first reaction is that within informal systems,
statements such as the above may be entirely suitable,
but doing any kind of formal reasoning on them is
extremely hazardous, given that the actual level of
modeling complexity *implicit in the language* is a
great deal higher than is explicit in the model.

As Jack noted, "cancer cells" is a simplistic label. We
immediately run into issues of semiotics, linguistics
and subject identity. IS_IN_PLACE_OF itself looks to be
a species of temporal, partitional, and mereotopological
(to use Smith's term [1]) ontological predicate, not to
mention all the background logics/ontologies that would
be necessary to support those, *before* we ever get to
the domains of biomedicine and cancer cells and how
that subdomain context would effect the use of the
predicate. The formal complexity is truly astounding,
if we're willing to address it. Starting with something
like McCarthy's [2] papers on modal logic it would be an
amazing feat to pull the necessary components together
to *accurately* create a logic to describe cancer cells
usurping lung cells -- a very interesting problem, to
be sure. These types of statements sound good to a human
ear (and indeed, humans are capable of dealing with a
great deal of cognitive dissonance and ambiguity) but
have almost no functionality in a machine ontology absent
the availability of that logic.

When attempting to then apply the abstract ontology in
developing an abstract model and a concrete, functional
system, it's extremely common to mistakenly intermix
the semantics of the ontology with that of the modeling
system, e.g., using Topic Map "scope" as an equivalent
to "context", where the former is a syntactic mechanism,
the latter requiring a full-blown ontology in its own
right (including its own academic conference series [3]).
In the "Semantic Web" community this is often taken to
the limits of plausibility in the use of Description
Logics wildly beyond their actual valid scope. Most people
using DL nowadays have no idea of those limits (and
seemingly have never done their homework enough to even
understand why that is the case, or deliberately avoid
confronting those limits as that would effectively end
the experiment in many cases).

Even something so simple as a "truth value" is quite a
bit more complicated in implementation, if one is trying
to be remotely accurate. Labeling something 'true' or
'false' is quite simple when one remains within the realm
of abstract or programmatic logic, but such notions of
truth are more appropriately modeled in the realm of
discourse, a huge ontological category that has yet to be
fully explicated. One of my [ex]advisors at KMi is doing
his research in the domain of discourse. He has little
to no background in epistemology; that does not trouble
him.

Most models are by necessity extreme simplifications of
reality; I just find that there are so many places where
modeling falls short that I tend to throw my hands up and
walk away. My own decision was to go entirely informal,
to "just let people talk", and not make any formal
ontological commitments (call me commitment-phobic). It's
my experience that such arguments as above are quickly
characterized as a form of uninformed whining, rather
than actually answering to the charges. Not being able to
answer the charge would be to admit to the ultimate toy-
nature of most of these systems (noting that DARPA is
funding a lot of "Semantic Web" research). And we need to
remember that these toys do actually get used, that there
is real danger inherent in mistakes, like stupidly
imprisoning someone because their name matches that of a
known terrorist -- talk about problems with subject
identity!

Alexander, this isn't particularly a rant on what you or
anyone is doing, more a caution on the oversimplification
of models and the overzealous belief in their validity
based on an ignorance of the real epistemological problems
inherent in the modeling process.

Steve Newcomb once characterized this as a great chasm,
and indeed, I've never yet seen it convincingly crossed
(and I include Cyc in that evaluation). I think what
Steve and others are doing now with the TMRM is trying to
address the shortcomings in Topic Maps in this area. I'm
not sure there is a solution, but then again, I am
admittedly a cynic when it comes to these issues. It
assuredly remains a very important and interesting area
of research.

Murray

[1] http://ontology.buffalo.edu/smith/
[2] http://www-formal.stanford.edu/jmc/
[3] http://context-07.ruc.dk/CONTEXT07MainPage.html
...........................................................................
Murray Altheim <murray06 at altheim.com>                              ===  = =
http://www.altheim.com/murray/                                     = =  ===
SGML Grease Monkey, Banjo Player, Wantanabe Zen Monk               = =  = =

        In the evening
        The rice leaves in the garden
        Rustle in the autumn wind
        That blows through my reed hut.  -- Minamoto no Tsunenobu