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

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

Hi, Sumit.

This is not a "bug" so much as a quirk of the machine-learned model used by the POS.  Currently, POS and Chunker make predictions in a pipelined fashion -- the Chunker uses POS tags as features when determining the label of a given chunk, but it also considers lexical features -- in this case, the evidence from the non-POS features appears to be strong enough to outweigh the incorrect signal from POS.

Regards,

Mark

---

From: illinois-ml-nlp-users-bounces AT cs.uiuc.edu [illinois-ml-nlp-users-bounces AT cs.uiuc.edu] on behalf of Sumit Das [sumit.das AT abzooba.com]
Sent: Wednesday, April 16, 2014 12:13 PM
To: illinois-ml-nlp-users AT cs.uiuc.edu
Subject: [Illinois-ml-nlp-users] Shallow Parse chunks -- Illinois Chunker

Hi,

I am using Illinois POS tagger and Illinois Chunker. For the following sentence the chunk output is fine but the POS coming is bit surprizing.

"I have an ipod mini and it is a great buy but I'm probably the only person that dislikes the itune software."

Chuck for dislikes is VP but POS for dislikes is NN

Can you please look into this.

Thanks & Regards

Sumit Das

Researcher
Abzooba India Infotech Pvt. Ltd.
Kolkata, India 

Phone No. : +91-9775193756

Email : sumit.das AT abzooba.com
http://cse.iitkgp.ac.in/~sumitd/