ctakes-dev mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From "Miller, Timothy" <Timothy.Mil...@childrens.harvard.edu>
Subject Re: head word identification
Date Mon, 02 Mar 2015 22:17:04 GMT
I've been using the DependencyUtility class with dependency parses.
There is a method getNominalHead(JCas, Annotation), that you can pass
the cas you are working with and the named entity whose head you want.
Assuming there are dependency parses in the pipeline it will give you
back a ConllDependencyNode object which is the head of the phrase. In my
uses it has been satisfactory though I haven't any rigorous checking.

On 03/02/2015 11:29 AM, Dligach, Dmitriy wrote:
> Hello,
> Is anybody aware of a reliable way of identifying the head word of a UMLS entity? In
the general domain, people often use Collins rules, but I’m not sure whether they would
be applicable to clinical entities.
> Until recently I was under impression that taking the last word of an entity would work
pretty well, but now that I have looked at the data more closely, I am not so sure. E.g. it
fails in these cases: “breast, left”, “ductal carcinoma in situ”, “carcinoma, consistent
with breast primary”.
> Dima
> Dmitriy (Dima) Dligach, Ph.D.
> Boston Children's Hospital and Harvard Medical School
> (617) 651-0397

View raw message