CoreNLP icon indicating copy to clipboard operation
CoreNLP copied to clipboard

Published 20 hours ago • verified publisher icon stanfordnlp

NER Tag Disambiguation Between "PERSON"/"PERCENT"

Open raletty opened this issue 9 years ago • 1 comments

In Mention.java, within the edu.stanford.nlp.hcoref.data package, I'm noticing some inconsistencies with how the ner types of "PERSON" and "PERCENT" are handled. As far as I could tell, this page details the various types possible for NER.

See below for examples of the mentioned inconsistencies within said Mention.java file:

  • Line 290: not sure where getGender is called from but "PERCENT" would be lumped into this case
  • Line 525: I don't believe "PER" by itself is a type (as attempted by the second condition)
  • Line 621/629: the case for "PERCENT" is never reached since it starts with "PER"
  • Line 798: "PERCENT" lumped in with "PERSON" in this case
  • Line 1294/1295: "PERCENT" lumped in with "PERSON" in this case

raletty avatar May 04 '16 19:05 raletty

These are great observations, thanks for bringing them up! I will investigate the situation further!

J38 avatar May 11 '16 08:05 J38