RE: Comments on the tension between bottom-up KR and use of top down ontologies

While I think there's a lot of iteresting issues in this space, let 
me point out that the mail below seems to think that the "OR" between 
top-down and bottom-up is an "XOR" - i.e. one of the other.  The 
vision I've been pushing for a long time is one where they both 
flourish, and even more importantly they link together into a Web of 
Semantic Definitions.   My very first briefs when the DAML language 
was being created included the idea that a key aspect of a 
"distributed ontological representation" was that it could combine 
top-down and bottom-up approaches - here's the words from a slide 
from 2000 - you can see I've been saying this for a long time now...
  -Jim H.
p.s. For US readers, this is one of the arguments that helped 
convince the govt to spend a chunk of your tax dollars on creating an 
ontology language :-)

(from DARPA presentation - ca. 2000 - apologies for formatting, 
copiy/pasted from a slide in ppt format)

Small communities define common semantics
Technical Vocabularies abound
Mission specific
Technical jargons
Shared values
Larger communities form around shared terms
Mapping and "articulation" become crucial
Interoperability at web languages level
Top-Down (organization defines critical aircraft properties)
          or bottom up (Oh, a "foxbat" is a Mig29)
Business case for improving communication!


At 11:14 AM -0500 1/20/07, Kashyap, Vipul wrote:
Bill,

I am glad you brought this up! This could make a good topic for a 
future BIONT/HCLSIG Telcon Agenda.

I do have another version of the Top-Down/Bottom-Up "tension" which 
brings up the same issue in a different context.

Top-Down: Use Cases => Ontologies => Mappings to Data => RDFize Data Sets
Bottom-Up: RDFize Data Sets => Ground RDF Graphs in 
Ontologies/Terminologies => See applicability to Use Cases.

Wherea, I do have a preference for one of the above, I do recognize 
the validity and appropriateness of the second
approach in various scenarios.

Am glad that we are having this debate as a community and if we 
leverage the discussions and thoughts proposed
around this, I think we would have made a contribution to the field.

Look forward to hearing from fellow HCLSIG-ers on this.

Cheers,

---Vipul

=======================================
Vipul Kashyap, Ph.D.
Senior Medical Informatician
Clinical Informatics R&D, Partners HealthCare System
Phone: (781)416-9254
Cell: (617)943-7120
<http://www.partners.org/cird/AboutUs.asp?cBox=Staff&stAb=vik>http://www.partners.org/cird/AboutUs.asp?cBox=Staff&stAb=vik

To keep up you need the right answers; to get ahead you need the 
right questions
---John Browning and Spencer Reiss, Wired 6.04.95

From: public-semweb-lifesci-request@w3.org 
[mailto:public-semweb-lifesci-request@w3.org] On Behalf Of William Bug
Sent: Saturday, January 20, 2007 10:02 AM
To: public-semweb-lifesci hcls
Subject: Comments on the tension between bottom-up KR and use of top 
down ontologies

Hi All,

This was recently posted to the UMLS list.

Given some of the issues we've been discussing, I thought others 
might appreciate some of the ideas recounted here by Gary Merrill 
from GlaxoSmithKline

I have my own take on this very very important issue, but I'd rather 
not editorialize on Gary's points - and give you a chance to process 
them as he so clearly expressed them.  Some familiarity with UMLS 
structure is helpful 
(<http://umlsinfo.nlm.nih.gov>http://umlsinfo.nlm.nih.gov).

By the way, a site relevant to our efforts is the Open Clinical site 
(KM for Medical Care 
- <http://www.openclinical.org/medTermUmls.html>http://www.openclinical.org/medTermUmls.html).

Cheers,
Bill

Begin forwarded message:


From: <mailto:gary.h.merrill@GSK.COM>gary.h.merrill@GSK.COM
Date: January 19, 2007 10:52:11 AM EST
To: <mailto:UMLSUSERS-L@LIST.NIH.GOV>UMLSUSERS-L@LIST.NIH.GOV
Subject: Re: MRHIER and AUIs
Reply-To: <mailto:gary.h.merrill@GSK.COM>gary.h.merrill@GSK.COM

William:

I think that was a very good non-techincal summary of some issues in the
Metathesaurus that can be difficult and confusing.  The nature and role of
AUIs (and their relationships to one another and to the CUIs that they
"realize") can require substantial thought.

I am always a little concerned when I see statements such as  "In an ideal
harmonious world,  NLM and all sources would agree, and Meta would become
a single unified principled
ontology."   I do not in fact think that this is necessarily true (under
some reasonable constraints it is in fact provably false), and definitely
do not think it should be taken as a disideratum.  Perhaps you do not
either, but I wanted to take this opportunity to say that, particularly in
the context of evolving empirical scientific theories, we should not
expect (and not necessarily even strive for) such a unified ontology.
(There are, of course, those who would disagree.)  The history of science
and the history of philosphy has shown the folly of this, and I would
argue that while striving for a certain "convergence" is desireable,
striving for the one true theory/ontology is not.  That's something of a
digression, but I take the strength of UMLS to lie in providing a way of
"communicating between" and using mulitple disparate (at times mutually
inconsistent) world views without imposing a strict ueber-ontology. Again,
there are those who tend to find the lack of the ueber-ontology to leave
them feeling insecure and adrift in metaphysical ream of uncertainty.

As I expressed to Chris in separate communication, from my perspective (as
a very application-oriented user), UMLS provides a usually adequate
representation of "concepts" (via CUIs), and terms/words/linguistic items
(via SUIs, LUIs, etc.).  What it does not provide a particularly crisp
representation of at the moment is "things" -- e.g., diseases rather than
disease names or disease concepts (that is, the extensional correlate of
the (intensional) concept/CUI).  AUIs are enlisted to support this to some
degree, but they are somewhat too closely allied to linguistic items
(terms) to carry the genuine semantic weight of "things" (extensions).  At
best, one ends up using sets of AUIs as equivalence classes to represent
the thing to which each of the AUIs "refers" (though "refer" here is, I
think, a bit misleading).  So in terms of a classic thing/word/concept
semantic hierarchy, my feeling is that UMLS does a good job of the
word/concept part, but the thing part is left a bit "mushy".  However,
there is room for substantial debate here, and many of the issues are
unclear.

Largely this is a consequence of construing UMLS as a -- surprise --
meta*thesaurus* rather than a meta*ontology*, and focusing on meaning
relations (e.g., synonomy) rather than more fundamental semantic relations
(e.g., denotation and extension).   I do have some ideas of how this might
be addressed, but won't even mention them here -- partly because working
them out requires substantial thought and care, and partly because I'm not
altogether sure of what the benefit would be (to most UMLS users) to
retrofitting such an approach to UMLS.

------------------------------
Gary H. Merrill, Director
Semantic Technologies Group
Statistical and Quantitative Sciences
GlaxoSmithKline Research and Development
Research Triangle Park, NC
919.483.8456


Bill Bug
Senior Research Analyst/Ontological Engineer

Laboratory for Bioimaging  & Anatomical Informatics
www.neuroterrain.org
Department of Neurobiology & Anatomy
Drexel University College of Medicine
2900 Queen Lane
Philadelphia, PA    19129
215 991 8430 (ph)
610 457 0443 (mobile)
215 843 9367 (fax)


Please Note: I now have a new email - 
<mailto:William.Bug@DrexelMed.edu>William.Bug@DrexelMed.edu





THE INFORMATION TRANSMITTED IN THIS ELECTRONIC COMMUNICATION IS 
INTENDED ONLY FOR THE PERSON OR ENTITY TO WHOM IT IS ADDRESSED AND 
MAY CONTAIN CONFIDENTIAL AND/OR PRIVILEGED MATERIAL. ANY REVIEW, 
RETRANSMISSION, DISSEMINATION OR OTHER USE OF OR TAKING OF ANY ACTION 
IN RELIANCE UPON, THIS INFORMATION BY PERSONS OR ENTITIES OTHER THAN 
THE INTENDED RECIPIENT IS PROHIBITED. IF YOU RECEIVED THIS 
INFORMATION IN ERROR, PLEASE CONTACT THE SENDER AND THE PRIVACY 
OFFICER, AND PROPERLY DISPOSE OF THIS INFORMATION.


-- 
Prof James Hendler				hendler@cs.rpi.edu
Tetherless World Constellation Chair		http://www.cs.umd.edu/~hendler
Computer Science Dept			301-405-2696 (work)
Rensselaer Polytechnic Inst			301-405-6707 (Fax)
Troy, NY 12180

Received on Sunday, 21 January 2007 01:12:28 UTC