Hello!
Basically I want to find a path between two NP tokens in the dependencies graph. However, I can't seem to find a good way to do this in the Stanford Parser. Any help?
Thank You Very Much
+2
A:
The Stanford Parser just returns a list of dependencies between word tokens. (We do this to avoid external library dependencies.) But if you want to manipulate the dependencies, you'll almost certainly want to put them in a graph data structure. We usually use jgrapht: http://jgrapht.sourceforge.net/
Christopher Manning
2010-05-02 04:12:34
Thank you.I'm actually using this to extract triples (Subject,Verb,Object) from sentences. My idea was to first detect the entities in the sentence, and then, for each pair of entities, find a path between them in the dependency graph that contained a verb. The problem is that, although this works for some cases, for some it doesn't. Like this one: "Paper is a thin, flat material produced by compressed fibers.". Here, the stanford parser finds a direct nsubj relation between "paper" and "material". Can you recommend any good way to do this triple extraction?Thank You.
pns
2010-05-02 10:53:38
In the SD representation, the copula (verb "to be") isn't treated as a predicate, rather the noun or adjective is treated as the predicate:thin(paper)material(paper).See the discussion in http://nlp.stanford.edu/pubs/dependencies-coling08.pdf (section 2.2)But if you wanted to undo this, you could look for a cop() dependency of the noun or adjective and rewrite it as be(paper, thin)be(paper, material)
Christopher Manning
2010-05-05 04:59:28