ctakes-dev mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From Lee Becker <lee.bec...@gmail.com>
Subject Re: cTAKES Dependency Parser / SRL updates
Date Fri, 05 Apr 2013 15:58:05 GMT
On Fri, Apr 5, 2013 at 9:26 AM, Chen, Pei <Pei.Chen@childrens.harvard.edu>wrote:

> I am planning to update the existing ctakes-dependency-parser ClearParser
> (which is no longer supported) implementation with Jinho's new ClearNLP
> parser and SRL. The models have been retrained with more data (MiPaq and
> SHARP) and much improved (1/4 memory footprint and faster).  Note that the
> current default pipeline actually has the SRL component commented out.
> I am planning to replace the existing code in trunk.  Please let me know
> if you have any objections or if you need the existing dependency parser
> code for some reason.  Otherwise, I would like to avoid supporting a
> dependency on a lib that is no longer supported with the same functionality.

Hi Pei,

If you need an example of how to prepare the data to go into ClearNLP and
how to take it back out, you can look at how ClearTK wraps ClearNLP.

This is the code currently in the master branch of ClearTK. Because it only
operates on the ClearTK typesystem it may be easier to read.

This is the code in the branch for issue-359 that is aiming to make the
wrappers more generic so that people can use their own type system, like
say cTAKES tokens and relationships.  These APIs are a work in progress,
but they might provide a useful path since cTAKES already depends on
ClearTK in parts, this might be a faster path as all you would need to do
would be to implement your own version of TokenOps, DependencyOps, and
SrlOps, and then you would create new classes that extend *_ImplBase.  I
would be curious to know if they are currently general enough to support a
type system like the one in cTAKES.

Let me know if you have any questions,

  • Unnamed multipart/alternative (inline, None, 0 bytes)
View raw message