Inconsistent results of POS tagging between core nlp demo and parser demo

问题

Inconsistent results of POS tagging between

P: http://nlp.stanford.edu:8080/parser/

and

C: http://nlp.stanford.edu:8080/corenlp/process

E.g.,

C: We went east/JJ to Oslo. P: We went east/RB to Oslo.

C: We are all/DT getting older. P: We are all/RB getting older.

C: Are you getting excited/VBN about your vacation? P: Are you getting excited/JJ about your vacation?

C: Did you do/VBP that? P: Did you do/VB that?

It seems that the parser performs better than core nlp, but I cannot replicate the parser results by switching between the model provided in the core nlp zip file.

Any idea?

回答1:

You will get different part of speech tag results if you use these different pipelines:

tokenize,ssplit,pos,lemma,parse

vs.

tokenize,ssplit,parse

The latter will perform part of speech tagging as part of the parsing process. The former uses the MEMM sequence tagging model that is dedicated to part of speech tagging.

来源：https://stackoverflow.com/questions/48939806/inconsistent-results-of-pos-tagging-between-core-nlp-demo-and-parser-demo