ctakes-dev mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From "Chen, Pei" <Pei.C...@childrens.harvard.edu>
Subject RE: srl bug with one word sentences
Date Fri, 12 Apr 2013 01:25:02 GMT
My vote would be for 2) Create a DependencyTreeHead which would make it explicit similar to
ClearTK's org.cleartk.syntax.dependency.type.TopDependencyNode.

Since there is already a temp fix, we can wait and see if there are any objections with adding
this type first?

From: Tim Miller [timothy.miller@childrens.harvard.edu]
Sent: Monday, April 08, 2013 3:54 PM
To: dev@ctakes.apache.org
Subject: srl bug with one word sentences

I filed a jira (https://issues.apache.org/jira/browse/CTAKES-184) for an
issue in the srl component of the dependency parser. Here is the issue:
The SRL component grabs the Dependency Node for each token using the
token span. When there is a one-word sentence there are two such nodes,
the "head" node which spans the sentence and the token node. There is a
quick fix checked in that looks at the "id" field, but it would be
preferable to fix it upstream so that users of the Dependency types
don't need to know about this issue.

Here are a few possibilities, any other ideas?
1) Change the span of the head node, maybe to 0 length at the start of
the sentence?
2) Create a separate type for DependencyTreeHead, perhaps spanless

Tim Miller, PhD
Postdoctoral Research Fellow
Children's Hospital Informatics Program
Boston Children's Hospital and Harvard Medical School

View raw message