illinois-ml-nlp-users AT lists.cs.illinois.edu

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

List archive

[Illinois-ml-nlp-users] Reg. identification of numbers using extended NER

From: Rakesh Guttikonda <rguttiko AT uwaterloo.ca>
To: illinois-ml-nlp-users AT cs.uiuc.edu
Subject: [Illinois-ml-nlp-users] Reg. identification of numbers using extended NER
Date: Tue, 27 Aug 2013 01:46:30 -0400
List-archive: <http://lists.cs.uiuc.edu/pipermail/illinois-ml-nlp-users/>
List-id: Support for users of CCG software <illinois-ml-nlp-users.cs.uiuc.edu>

Hi Everyone,

I am using the extended NER tagger (v2.3) for tagging the text in a project. However, I have noticed that the NER tagger was not able to properly identify the numbers (CARDINAL type) in some sample text.

For example, I have tried running the "rundemo.sh" bash script which tags the text in "longparagraph.txt". In the output of tagged text, the following numbers have not been identified: "...:20 first-graders ..." or "....six staff members..." or "...23 executive..." etc. Infact, I did not see any CARDINAL tagged number which seems weird.

However, when I tagged the same text using the online demo version of extended NER (http://cogcomp.cs.illinois.edu/demo/ner-extended/?id=28), it identified all the numbers.

So, I rechecked the config settings in ontonotes.config, but everything seems to be perfect. The DATE, TIME, PER, ORG, GPE, MONEY seem to be correctly tagged and only numbers seem to be missing.

How is that the online demo is able to identify the numbers and not the downloaded software? Is there any setting that I am missing? Please note that I have downloaded the latest version of extended NER (v 2.3).

Any help in this matter is much appreciated.

Thanks,

Rakesh

[Illinois-ml-nlp-users] Reg. identification of numbers using extended NER, Rakesh Guttikonda, 08/27/2013