Search code examples
nlpstanford-nlpopennlp

extract NP-VP-NP from Stanford dependency parse tree

I need to extract triplets of the form NP-VP-NP from the dependency parse tree produced as the output of lexalized parsing in Stanford Parser.

Whats the best way to do this. e.g. If the parse tree is as follows:

(ROOT
  (S
    (S
      (NP (NNP Exercise))
      (VP (VBZ reduces)
        (NP (NN stress)))
      (. .))
    (NP (JJ Regular) (NN exercise))
    (VP (VBZ maintains)
      (NP (JJ mental) (NN fitness)))
    (. .)))

I need to extract 2 triplets:

  1. Exercise-reduces-stress and
  2. Regular Exercise-maintains-mental fitness

Any ideas?

Solution

  • There are two natural options here. One is to run Semgrex over the dependency tree (side note: what you have in the question is a constituency tree), with a pattern like:

    {pos:/V.*/}=verb >/.subj.*/ {}=subject >/.obj/ {}=object

    Another option is to use the Stanford Open IE system. This will give you a more broad semantics of '(subject; relation; object)' triples, where the relation does not have to be a verb.