CODEWITHSUNDEEP
pythonparsingnlpnltkstanford-nlp
Yarik
Yarik

Reputation: 81

Using StanfordParser to get typed dependencies from a parsed sentence

Using NLTK's StanfordParser, I can parse a sentence like this:

os.environ['STANFORD_PARSER'] = 'C:\jars' 
os.environ['STANFORD_MODELS'] = 'C:\jars'  
os.environ['JAVAHOME'] ='C:\ProgramData\Oracle\Java\javapath' 
parser = stanford.StanfordParser(model_path="C:\jars\englishPCFG.ser.gz")
sentences = parser.parse(("bring me a red ball",)) 
for sentence in sentences:
    sentence

The result is:

Tree('ROOT', [Tree('S', [Tree('VP', [Tree('VB', ['Bring']),
Tree('NP', [Tree('DT', ['a']), Tree('NN', ['red'])]), Tree('NP',
[Tree('NN', ['ball'])])]), Tree('.', ['.'])])])

How can I use the Stanford parser to get typed dependencies in addition to the above graph? Something like:

  1. root(ROOT-0, bring-1)
  2. iobj(bring-1, me-2)
  3. det(ball-5, a-3)
  4. amod(ball-5, red-4)
  5. dobj(bring-1, ball-5)

Upvotes: 8

Views: 2064

Answers (1)

dmcc
dmcc

Reputation: 2529

NLTK's StanfordParser module doesn't (currently) wrap the tree to Stanford Dependencies conversion code. You can use my library PyStanfordDependencies, which wraps the dependency converter.

If nltk_tree is sentence from the question's code snippet, then this works:

#!/usr/bin/python3
import StanfordDependencies

# Use str() to convert the NLTK tree to Penn Treebank format
penn_treebank_tree = str(nltk_tree) 

sd = StanfordDependencies.get_instance(jar_filename='point to Stanford Parser JAR file')
converted_tree = sd.convert_tree(penn_treebank_tree)

# Print Typed Dependencies
for node in converted_tree:
    print('{}({}-{},{}-{})'.format(
            node.deprel,
            converted_tree[node.head - 1].form if node.head != 0 else 'ROOT',
            node.head,
            node.form,
            node.index))

Upvotes: 6

Related Questions