Hi Barry, I just wanted to drop a quick note that I appreciate
your visit and the inspiring debate. It's not like I am ignoring
your criticism. I have a sceprical approach to the feasibility
of your grand vision of the great all-unifying ontology, which is
of the same origin as my scepticism against the need to pound on
the realism point. In the end, whether we might be optimistic or
sceptical about what humankind will be able to know, what matters
is what we do today in the world where theories are still
developing and shifting.
And I agree with many points regarding the quality we should
strive for when building these things such as a Reference Information
Model or ontologies. Although I do not necessarily buy every
specific solution as to how to do things differently. It might
be neither in your or in my personality to focus on agreement
(if imperfect) rather than harp on the disagreement, but let me
just try an honest attempt:
Your top-level ontological categories to distinguish are:
1) continuant vs. occurrant
2) independent vs. dependent
3) universal - particular - individual
I agree with that split, and I want to think the RIM agrees with
it, although it may look otherwise to you.
If you can -- for a moment -- resolve the sticky point about the
Information model vs. model of the "real world", and if you buy
that the RIM is a model of Information about the world, and therefore
if you implicitly prefix every name of a RIM class C with the
phrase "Information about C", then you might appreciate the
HL7 RIM Class Barry Smith's Ontological Categories
Entity continuant, independent
Role continuant, dependent
Participation occurrant, dependent
Act-Information occurrant, independent
ActRelationship occurrant, dependent
You probably know that we are dealing with the dimension of
universal - particular - individual using the Act.moodCode and
the Entity.determinerCode. It may look strange to you, and it
is not perfect, but we have our reasons, and did not regret this
choice. We can discuss that elsewhere.
Just as you are now, so have we been very interested in
laying down a rich set of well defined relationships, and our
large variety of ActRelationship.typeCode, Role.classCode
(i.e., Entity-Relationship) types are a great step forward
in a context where people didn't even think it mattered to
distinguish a "reason" link from a "composition" link. I hope
you appreciate that as a step forward from where we started
in 1999. Now, when you look at those you find many oddballs,
which is what people have thrown into the mix haphazardly
in the hurry of getting things done. So, don't just pick the
first oddball and declare the whole thing for rubbish please.
We have some contributions to make even in your zoology of
part-like relationships (which are rather simple relationships
to define and think about). We distinguish at least:
which are relevant distinctions that our colleagues thought
were too "philosophical" to even bother with. These distinctions
are not even always made in other ontologies, and certainly not
in the year 1999 when we introduced these.
I think we are missing a notion of
as opposed to part and member. But have not had the ability to
make the case and show the repercussions of adding this relationship.
We also introduced a common way of quantifying all material
relationships, which is very simple and solves a problem which
has been a subject of great confusion in all ontologies where
such quantification is important (notably all drug "terminologies").
All that does not claim that the definitions couldn't be cleaned
up or made more rigorous, but the distinctions we make have merit
and the way we set up the quantification of these roles work for
what we have been trying to do.
We also introduced the distinction between Participation (and their
types) and Roles, and closed the loop between Role and the scoper,
which I think are novel, significant, innovative and true.
That all does not claim that there are not rough edges, even deep
logical problems that we might get into. But to clear those, we
need to go from a *specific* problem analysis to a *specific*
solution that considers the rationale for the things we have, some
of it documented some passed down only in oral tradition and in
tacit assumptions or common hermeneutic frames of reference
(something tells me that you won't appreciate the latter, but
thought it would fit your use of the word "exegesis" this
It was enlightening to see that even you are not exempt from
producing models you do not totally agree with yourself about.
Such as how to model "food fragments in vomitus." We are loaded
with these issues, which are practical compromises to go forward
even though we may not like it all perfectly.
So, I look forward to more of your constructive criticism in the
spirit of making a good thing better, even shedding off the things
that truly are misguided (see, even I can be a realist sometimes .
But they have to be proven to be misguided and a feasible
alternative has to fill the void that the criticism would create.
I'll be glad to put your specific change proposals forward in
the process of HL7, and will obviously strongly support what I
can see as an improvement.
Two more details:
1) you have trouble interpreting what is meant by saying that
an Act is both the "Performance and its Documentation."
As a implicit principle you have to take that the RIM always
models Information-about Act, Entity, etc. And even once we
realize that we do not model "acts" but "information about acts,"
the explicit identification of "Information about Act" and
"Documentation" is still valid, as it speaks to a perpetual
problem that we have in the discussion of how to construct a
good information model about our domain.
The problem is that there are two very oposed views on how to
model healthcare information. There is one group who thinks
healthcare information is a collection of documents
or "reports", and that modeling healthcare information therefore
means to model "reports".
As opposed to this, many people in HL7 know that we model
information about the healthcare services (acts) which may or
may not be found in a report document, but we believe that
this information about Acts needs to be reflected in the model
based on an analysis of what these Acts are (in "reality" if
you will) and independent of how these may be talked about in
conventional text documents. So, we say, we model the "information"
not the "report". This is the dispute that is behind the statement
that an "Act is both the Performance and its Documentation", it
means that we claim that the RIM Acts model information about
healthcare services, and that we see text documents or hierarchical
arrangements of this information to be a matter of "rendering"
this information in documents and folders etc. and we reject to
have document issues creap into the analysis of Acts in the real
world on which we want to base our model of information about
acts. For instance, we believe that a "reason" is a link between
two Act statements, and that it is not appropriate to model
"reason" simply as a "section of a document" or a mere
We are claiming that the information models that people use
to control acts in workflow-automation should be the same as the
model people use to document acts after they are done with the
work, and that should be the same model as people use who
document historical acts that occurred far in the past. So, in
a sense we are trying to be far more "realists" than those people
who think that all we should care is to model "record structures"
about regardless whether these record structures talk about
acts or carrots in vomitus.
Don't know if that helped to reach out.
2) I am puzzled by your issue regarding the symmetry of
the "adjacency" relation.
Clearly on the instance level, adjacency is symmetric, but
why do we have to conclude that adjacency must therefore
be symmetric on the universal level?
It seems to me that the situation can be described in the
Let R be a relation:
R : (A union B) x (A union B)
which means that
some a in A, b in B satisfies b R a
Why are you surprized that just because the following
axioms are true
forall a in A : a R b => b in B
forall b in B : b R a => a in A
which lead you to think that the relation is symmetric,
that it then does not follow that
forall b in B : some a in A : b R a?
some Cytoplasma is-adjacent-to Nucleus
and Nucleus is-adjacent-to some Cytoplasma
does not mean that
all Cytoplasma is-adjacent-to Nucleus.
This problem is reminescent of what the S-E-P triplets are
addressing. The notion of "Cytoplasma" is a homophone, as it
S) cytoplasmatic matter
E) the entirity of the cytoplasma of a cell, or
P) a part or portion of cytoplasma
So, while it may be said that in nucleated cells, the
entirity of cytoplasma is adjacent to the nucleus, it is
certainly not true to say that every portion of cytoplasma
is adjacent to the nucleus, nor does it even make sense
to say that "cytoplasmatic structures" have any adjacency
Also, of course there are both nuclei that are not adjacent
to cytoplasma and there are denucleated cells which have
cytoplasma which has no adjacency to a nucleus at all times
Is this touching your issue at all? I couldn't quite see
how this would be an example for the sort of logical problems
that Rutherford and Goedel went into with the logical
foundation of mathmatics. I actually agree that your project
will have to answer to the problem that not even Rutherford
can solve for Mathematics, but I didn't follow your example
as even getting near the illustration of such a problem?
So, thanks again for your visit, the inspiring discussions,
and sorry for the long email about this large pile of things
that have accumulated.
Gunther Schadow, M.D., Ph.D. firstname.lastname@example.org
Associate Professor Indiana University School of Informatics
Regenstrief Institute, Inc. Indiana University School of Medicine