natural icon indicating copy to clipboard operation
natural copied to clipboard

Published 20 hours ago • verified publisher icon NaturalNode

POS Tagger accuracy?

Open giorgio79 opened this issue 7 years ago • 4 comments

Do you have data on POS tagging performance?

  • en-pos is 96.43% https://github.com/FinNLP/en-pos
  • Stanford POS is around 97% https://nlp.stanford.edu/~manning/papers/CICLing2011-manning-tagging.pdf

giorgio79 avatar May 23 '18 09:05 giorgio79

No, but we can run it on a corpus to see how it performs. Do you have a suggestion for such a data set?

Hugo

Hugo-ter-Doest avatar May 23 '18 21:05 Hugo-ter-Doest

Good question. Never ran tests myself, but the test collections are referenced here https://aclweb.org/aclwiki/POS_Tagging_(State_of_the_art)#Test_collections

giorgio79 avatar May 29 '18 14:05 giorgio79

There are problematic tags (digital marketing is greatwill give marketing as a verb) but in a large corpus, it should not matter too much. In smaller texts, it will be problematic.

lazharichir avatar Nov 20 '18 22:11 lazharichir

@giorgio79 did you ever get any updates on this? I'd like to also use an accurate pos tagger.

ariel-frischer avatar Jun 23 '20 23:06 ariel-frischer