Extracting noun phrases

Even though, spaCy's rule-based noun chunking facilities are not yet supported, you can still extract noun phrases using the Berkeley Neural Parser (benepar).

Install dependencies

First you'll need to install the tool by issuing pip install benepar or pip install huspacy[np]

Usage

Then, benepar models should be downloaded and added to a HuSpaCy pipeline:

import benepar
import os

# Workaround for incompatible protobuf versions
os.environ["PROTOCOL_BUFFERS_PYTHON_IMPLEMENTATION"]="python"

benepar.download('benepar_hu2')

nlp.add_pipe("benepar", config={"model": "benepar_hu2"})

One can use this simple method to extract maximal noun phrase:

from spacy.tokens import Span
from typing import *

def extract_max_np(span: Span) -> Iterable[Span]:
  if "NP" in span._.labels:
    yield span
  else:
    for child in span._.children:
      yield from extract_max_np(child)

Then

doc = nlp("Ők korábban népszavazási kérdéseket jelentettek be, és azt ígérik, folytatják.")

for sent in doc.sents:
  for np in extract_max_np(sent):
    print(np)

prints:

Ők
népszavazási kérdéseket
azt