On Sep 14, 2006, at 10:36 AM, Drew McDermott wrote:
[Chimezie Ogbuji]
Seems to me the biggest barrier is in coming to a consensus on
an appropriate placeholder vocabulary and not neccessarily on
determining
all the various ways in which a person (and their related data)
could be
expressed in a patient record.
I'm not sure I'm agreeing or disagreeing with Chimezie (it depends on
what's meant by "consensus" here), but I'd like to emphasize a point
others have made in this discussion: Deciding to use a particular
ontology is not like deciding who to marry. It's just a _vocabulary_.
This is different from most definitions of ontology I am aware of. I
think confusing ontology with vocabulary will get us into trouble.
You can mix and match terms from different vocabularies, and you can
make up a new class that's the intersection of two classes from
different ontologies.
Of course you don't want to (a) misuse an ontology; (b) duplicate what
someone else has already done. But it seems to me that if FOAF
includes a term for Person (defined as broadly as Dan Brickley says it
is),
This definition seemed to be along the same lines as the one in the
HL7 RIM, for which a criticism was recently posted on this list
then everyone in the world might as well use foaf:Person when
they want to talk about persons. If a person in a medical context has
special properties, then subclass foaf:Person. If you discover at
some point that you (or HL7 RIM) has produced a class equivalent to
one defined previously, then declare them equivalent (owl:sameAs).
Can two classes be declared identical if both their definitions are
incoherent?
[It seems to me that it would be good practice to use the earliest
defined term for something, partly out of courtesy and partly so
everyone will converge on the same term, but that's an orthogonal
issue.]
Why should courtesy have anything to do with it? Surely everyone
should converge on the best definition, not the first? Why is the
first especially deserving of courtesy?
I think FOAF is an ideal application ontology for its stated goals -
linking communities of people together, albeit with a bias towards
computer science types.
I don't think it is appropriate for biomedical science, and I don't
think your recommended practice of subclassing foaf:Person is a good
idea. This will lead to a hodge-podge of non-disjoint classes. Others
have recommended of separating a Person out from the roles a Person
participates in - surely this is better?
I don't see the problem here as being particularly difficult - there
are continuants, and there are biological continuants such as
organisms. These continuants participate in different processes,
taking on different roles at different times. See [1]
Aside from the very serious problem of lack of formal definitions for
the central classes Person and Agent, there are a number of problems
in using FOAF for science
Many of the properties such as foaf:geekcode are relevant for a very
small proportion of human beings. Surely these should be separated
out in a separate ontology of
FOAFCharacteristicsRelevantToMaleComputerScienceUndergrads, in
accordance with well-accepted principles of normalised design?
Here is something illustrative of the problem with FOAF; from the
FOAF documention
http://xmlns.com/foaf/0.1/#term_dnaChecksum
"foaf:dnaChecksum -- A checksum for the DNA of some thing. Joke."
Whilst it is surely tempting to litter our ontologies with jokes -
foaf:motherInLaw jokes, ChickenCrossingRoad in ontologies of animal
behaviour, reasoning about the odor receptors of DogsWithNoNose and
so on - I'm afraid I don't think jokes have any place in a reference
ontology. Sorry to be so humorless. I'm sure the designers of GALEN
regret adding "vomitus has_part carrot" to their ontology, after a
certain ontologist uses this as a frequent point of criticism. I'm
inclined to agree - in-jokes will cause problems for both human and
computer users.
foaf then goes on to reverse its position, and states: "The
foaf:dnaChecksum property is mostly a joke, but also a reminder that
there will be lots of different identifying properties for people,
some of which we might find disturbing."
...perhaps invoking a distant Gattaca-like dystopian future.
well obviously this is no joke in biomedical science where DNA
checksums are commonly applied to database records, and a wealth of
genotypic data is available for human and other organisms, and with
individual genome resequencing around the corner. Obviously a per-
human DNA checksum is a naive notion, but it's hardly inconceivable
that we'd want to record euchromatic DNA sequences for samples of
cells on a per human basis. This would have to be represented
appropriately, to make sure the genotypes of cancerous cells are not
mixed up with normal cells. The presence of a half-jokey half-serious
dnaChecksum property is highly inappropriate and will add confusion.
How about a deal - we'll let FOAF define properties within the FOAF
designers realm of expertise, like listing the different kinds of
online chat accounts, and you leave the DNA stuff to us.
If there are aspects of FOAF that are of use to biomedical science
(I'm not sure what these are), then these should be separated out
into a minimal ontology. If people want to reason over databases to
determine if genotypes correlate with foaf:OnlineGamingAccount then
they can do so by linking the appropriate ontologies, but
foaf:jabberID must be strictly separated from ontologies for doing
science.
Sorry if this is discourteous but I really find it bizarre that FOAF
is even being considered here.
Chris
[1] http://www.websemanticsjournal.org/ps/pub/2006-13
@ARTICLE{Smith2006,
author = {Smith, B. and Ceusters, W.},
title = {Referent Tracking for Treatment Optimisation in
Schizophrenic Patients},
journal = {Journal of Web Semantics},
year = {2006},
volume = {4},
number = {3},
abstract = {
The IPAP Schizophrenia Algorithm was originally designed in the form
of a flow chart to help physicians optimise the treatment of
schizophrenic patients. We examined the current version from the
perspective of recent work on terminologies and ontologies thereby
drawing on the resources of Basic Formal Ontology, and this with the
objective to make the algorithm appropriate for Semantic Web
applications. We found that Basic Formal Ontology is a rich enough
theory to represent all the entities involved and that applying the
theory to the IPAP schizophrenia algorithm results in a
representation that can be used by software agents to perform
monitoring and control in a referent tracking environment.
},
}
--
-- Drew McDermott
Yale Computer
Science Department