CODEWITHSUNDEEP
pythonnlpspacyhuggingface-transformersdependency-parsing
aravind kamarsu
aravind kamarsu

Reputation: 77

Extract subtree (Phrase) for nsubj elements using spacy

The aim is to extract the sub-tree (phrases) from the sentence if the 'nsubj' exists in the given sentence.

Here is the code which I am using:

import spacy    
nlp = spacy.load('en')
piano_doc = nlp('The alarm clock is, to many high school students, a wailing monstrosity whose purpose is to torture all who are sleep-deprived')
    for token in piano_doc:
        if token.dep_ == 'nsubj':    
            print (token.text, token.tag_, token.head.text, token.dep_)
            subtree = token.subtree
            print([(t.text) for t in subtree])
            print('*' * 50)

The output we get is: clock NN is nsubj

['The', 'alarm', 'clock']

purpose NN is nsubj

['whose', 'purpose']

who WP are nsubj

['who']

But the output i am expecting in the case of nsubj is the whole subtree i.e.

purpose NN is nsubj

['whose', 'purpose','is','to','torture']

who WP are nsubj

['who' ,'are' ,'sleep-deprived']

Upvotes: 1

Views: 2342

Answers (1)

polm23
polm23

Reputation: 15633

As krisograbek mentioned, your understanding of a subtree is not what a subtree is in spaCy, or in dependency parsing in general.

In dependency parsing, if you have a subject and a verb, the verb is the head. This means the subtree of the subject does not include the verb.

I am not sure exactly what you want but maybe you should try token.head.subtree for the subject.

Upvotes: 1

Related Questions