Support for users of CCG software closed 7-27-20

["Sammons, Mark" <mssammon AT illinois.edu>]

I’m running some local tests to verify the observed pos behavior.

TextAnnotations are now created by TextAnnotationBuilder subclasses. Character offsets are central to the CCG representation of text and annotations and will remain so.

 

Thanks,

Mark

 

 

From: Daniel Khashabi [mailto:khashab2 AT illinois.edu]
Sent: Saturday, September 10, 2016 3:52 PM
To: Richard Eckart de Castilho
Cc: illinois-ml-nlp-users AT lists.cs.illinois.edu
Subject: Re: [[Illinois-ml-nlp-users] ] Trouble upgrading to 3.0.71

 

Regarding the POS issues: that sounds a little unlikely. 

We have a unit-test to ensure the quality is kept between changes: https://github.com/IllinoisCogComp/illinois-cogcomp-nlp/blob/master/pos/src/test/java/edu/illinois/cs/cogcomp/pos/tests/TestDiff.java 

Could you compare your usage with this unit test? 

 

Char offset: not sure; Mark should comment. 

 

Daniel 

 

 

 

On Sat, Sep 10, 2016 at 3:19 PM, Richard Eckart de Castilho <richard.eckart AT gmail.com> wrote:

Hi all,

I have some trouble upgrading from 3.0.44 to 3.0.71.

After changing my code to disable lazy loading and replacing
calls to getView(...) with addView(...) the code compiles
nicely again. However, the POS tagger is producing bad results
now.

With 3.0.44, "This is a test ." is tagged as

  "DT",   "VBZ", "DT",  "NN",   "."

With 3.0.71, I get

  "NNP", "NN", "NN", "NN", "NN"

Maybe something fundamental in the data model has changed.
Any ideas what that might be?

I noticed that in TextAnnotation, the setTokens() method
has been deprecated. It took the character offsets. The
new TokenLabelView doesn't seem to bear character offsets.
Are you going to drop support for character offsets in the
future?

Cheers,

-- Richard