John, (01)
>I agree with everything that Kathy said up to the
>following sentence:
>
> > I think ontologies need to be able to represent
> > probabilistic relationships like these.
>
>I certainly agree that such relationships should be
>represented somewhere, but the question of where
>they should be represented is another issue.
>
>For years, I have maintained that the base ontology
>(i.e., what is usually called the "upper level")
>should have very few detailed axioms of any kind.
>
>In fact, I would be delighted to have a base ontology
>that contains nothing but a type hierarchy plus the
>minimal constraints that everybody can agree to without
>reservation. (02)
I would hope you agree that there are conditions (e.g., diseases,
faults, crimes ...) that we could like to diagnose. I hope you agree
that conditions have symptoms, and there are tests for conditions. I
hope you agree that symptoms and tests are examples of something more
general, which I'll call evidence. I hope you agree that Boolean
evidence for Boolean hypotheses has (possibly context-dependent)
sensitivity and specificity. (By "Boolean" evidence, I mean evidence
that can have two possible values, which I'll label "positive" and
"negative", where "positive" is evidential to its related hypothesis
being true and "negative" is evidential to its related hypothesis
being false.) (03)
>I recommend that all other information be
>included in collections of axioms of the kind that Cyc
>calls "microtheories". (04)
If ontologies contain only everything that everybody can agree on
without reservation, and everything else goes into microtheories,
then I'm afraid our ontologies are going to be empty, and what we
used to call ontologies will have to be relabeled as microtheories. (05)
SUMO hasn't been officially adopted yet, has it? (06)
>As a general principle, all knowledge derived from
>observation should be put in the microtheories, (07)
Um... Can you please explain to me what knowledge is NOT derived
from observation? (08)
>and
>only the definitional information that is derived
>from the most basic conventions for distinguishing
>one type from another should be included in the base
>ontology. (09)
So you don't want a medical ontology to represent that medical tests
have sensitivity and specificity? That's pretty basic, IMHO. (010)
>In general, it is far better to leave the
>base ontology underspecified than to include any axioms
>that are controversial or likely to be revised and
>updated as more information becomes available. (011)
I don't think it's at all controversial to assert that medical tests
have sensitivity and specificity, or that conditions have base rates. (012)
Whether we put specific base rates of specific conditions, or
specific sensitivities and specificities of specific tests, into an
ontology is a different question. (013)
>If you want to call those microtheories part of some
>larger specification, which might even be called an
>"extended ontology", I have no objection. (014)
Definitions of medical conditions (sometimes asserted as axioms in
medical ontologies) often consist of listing clusters of symptoms and
tests that are characteristic of the condition. In many such cases,
it would be more accurate from a medical standpoint to define the
condition by means of a probability model in which presence of the
condition is a hidden unobservable variable whose probability is
inferred from observable indicators. That would require
probabilistic axioms in the ontology or microtheory or extended
ontology or whatever we are going to call it. (015)
So what should the line be between ontology and microtheory? I think
this should not be legislated a priori on purely philosophical
grounds. I think it should be something the community collectively
decides, based partly on philosophical clarity, but also based on
what distinctions turn out to be useful in practice. Of one thing
I'm pretty sure -- the line should not be drawn arbitrarily to
exclude anything probabilistic because we have legislated that
ontologies can't be probabilistic. (016)
>But it's essential to distinguish observational information,
>which is constantly being updated and revised, from the
>classifications in the base (or upper level) ontology. (017)
If we change your words to, "...it's essential to distinguish
observational information that is constantly being updated and
revised..." then I agree wholeheartedly. (018)
In the example I gave, the information about the individual patient
who had a 1/1000 chance of having a condition, which increased to 4%
after she began complaining of a symptom of the condition, and then
increased to 80% after she was given a test that came up positive,
should NOT be part of the ontology. (019)
Do you think the knowledge that C is a medical condition should go
into a medical ontology? Do you think the knowledge that S is a
symptom characteristic of C should go into a medical ontology? That
T is a test for C? (All this knowledge was obtained from observation,
I might note.) (020)
If the answer to those questions is no, then I think your use of the
term ontology differs greatly from the use it is coming to have in
the practice of ontological engineering. (021)
If the answer is yes, then what is the justification for refusing to
include the sensitivity and specificity of T in the ontology, in
cases in which the sensitivity and specificity are extremely
well-characterized, and agreed upon by (almost) everyone (almost)
without reservation? (022)
Kathy (023)
_________________________________________________________________
Message Archives: http://ontolog.cim3.net/forum/ontolog-forum/
Subscribe/Config: http://ontolog.cim3.net/mailman/listinfo/ontolog-forum/
Unsubscribe: mailto:ontolog-forum-leave@xxxxxxxxxxxxxxxx
Shared Files: http://ontolog.cim3.net/file/
Community Wiki: http://ontolog.cim3.net/wiki/
To Post: mailto:ontolog-forum@xxxxxxxxxxxxxxxx (024)
|