Hello Ian,

Thursday, August 17, 2006, 8:49:01 PM, you wrote:

IC> The concrete thinking problem is "is machine processable, is good".

IC> Computer languages in use are max 40 years old. SNOMED is 32 years 
IC> old. The oldest practically useful computer is how old - <20 years 
IC> perhaps. English language is rather older than modern western 
IC> medicine. Strangely, we can still understand English written a couple 
IC> of hundred years ago.

IC> Modern studies on coding demonstrate considerable intercoder 
IC> variability, even amongst experts.

IC> It seems likely to me that computers will conquer natural language 
IC> processing faster than clinicians en masse can navigate a controlled 
IC> vocabulary with accuracy and speed.

I am very fascinated by grammar and its very powerful in very
controlled domains, as in programming languages, but the English
language is not controlled and I really doubt this will happen.
Parsing simple sentences leads to quite large parse trees and its very
hard for the computer to pick the correct one.

eg "The man boarded the plane with a suitcase"

now who has the suitcase, the man or the plane, we know its the man,
but the computer has a lot of trouble with this simple statement. Its
solved by probability analysis and frequency analysis, but trying to
make sense of notes about a complex problem coded with the average
clinicians terse expressions is a big ask. Especially when mistakes
are potentially costly.

I would think that unless we can "use" computer processable coding
there is little point in getting terms out of text anyway, as you are
saying there is no use for them.

google "semantic web" - its possible everyone else is wrong and we
will never have computer processable medical records, but I doubt it.
If its going to work then constraining/assisting input using the
smarts in the terminology (and using this info usefully )needs to be
proven first. If that works then natural language processing is
another layer on top, and not a substitute.


"Using Snomed-CT" should be invisible to clinicians, and natural
language processing is not the only way to this goal.


-- 
Best regards,
 Andrew                            mailto:[EMAIL PROTECTED]

Andrew McIntyre
Buderim Gastroenterology Centre
www.buderimgastro.com.au
PH: 07 54455055 FAX: 54455047

_______________________________________________
Gpcg_talk mailing list
[email protected]
http://ozdocit.org/cgi-bin/mailman/listinfo/gpcg_talk

Reply via email to