Re: [Gnumed-devel] need assessment from fellow clinicians

[Karsten Hilbert]

> >You'll get into a quagmire that decades of natural language
> >processing and AI have failed to make easy even with
> >specialized vocabularies/ontologies as in medicine.
> I realise that : the idea is that the user configures the parser to his 
> particular idiosyncratic style of
> rushed note entry during consultation,
Well, Richard's concept of edit area fields + phrase wheel do
just that. Except that Richard helps the natural language
parser by indicating "this ought to be plan text" via using an
input field for "plan". Then the phrasewheel kicks in
contextually suggesting previously used input text. This is
likely as powerful as we'll need. No *true* NLP required.

> Another thing that might be useful here is hidden markov model training 
> ;  as the user corrects the tentative
> tree updates, a hmm is trained to make a better parse next time . The 
> problem here might be ,
> hmm are being used to classify only 2 categories in practical use for 
> spam nowadays, but can gnumed
> use it to classify multiple categories ?
Clearly so but that's where we'll get into research instead of
getting things done. That's exactly the reason why hardly
anything of practical value ever leaves a (German) university.

Karsten
-- 
GPG key ID E4071346 @ wwwkeys.pgp.net
E167 67FD A291 2BEA 73BD  4537 78B9 A9F9 E407 1346