Hello,
We recently updated cTAKES (http://incubator.apache.org/ctakes/) to
openNLP 1.5, including retraining a parser wrapping the opennlp parser
on our clinical treebank. I noticed when running that the parser no
longer outputs function tags. For example, I get this output with 1.4:
(TOP (S (NP-SBJ (PRP She)) (VP (MD will) (VP (VB follow) (PRT (RP
up)) (PP (IN with) (NP (NN ENT))))) (. .)))
whereas I get this output with 1.5:
(TOP (S (NP (PRP She)) (VP (MD will) (VP (VB follow) (PRT (RP up))
(PP (IN with) (NP (NN ENT))))) (. .)))
(Note the first has NP-SBJ for the subject but the second doesn't.)
These are trained on the same training data. Does anyone know what may
have changed between versions? The tags are not essential but they are
in the training data and we had some downstream components that made use
of them as features.
--
Tim Miller, PhD
Postdoctoral Research Fellow
Children's Hospital Informatics Program
Boston Children's Hospital and Harvard Medical School
617-919-1223
