pipeline/nltk/parse/corenlp.py

# Natural Language Toolkit: Interface to the CoreNLP REST API.
#
# Copyright (C) 2001-2023 NLTK Project
# Author: Dmitrijs Milajevs <[email protected]>
#
# URL: <https://www.nltk.org/>
# For license information, see LICENSE.TXT

import json
import os  # required for doctests
import re
import socket
import time
from typing import List, Tuple

from nltk.internals import _java_options, config_java, find_jar_iter, java
from nltk.parse.api import ParserI
from nltk.parse.dependencygraph import DependencyGraph
from nltk.tag.api import TaggerI
from nltk.tokenize.api import TokenizerI
from nltk.tree import Tree

_stanford_url = "https://stanfordnlp.github.io/CoreNLP/"


class CoreNLPServerError(EnvironmentError):
    """Exceptions associated with the Core NLP server."""


def try_port(port=0):
    sock = socket.socket(socket.AF_INET, socket.SOCK_STREAM)
    sock.bind(("", port))

    p = sock.getsockname()[1]
    sock.close()

    return p


class CoreNLPServer:

    _MODEL_JAR_PATTERN = r"stanford-corenlp-(\d+)\.(\d+)\.(\d+)-models\.jar"
    _JAR = r"stanford-corenlp-(\d+)\.(\d+)\.(\d+)\.jar"

    def __init__(
        self,
        path_to_jar=None,
        path_to_models_jar=None,
        verbose=False,
        java_options=None,
        corenlp_options=None,
        port=None,
    ):

        if corenlp_options is None:
            corenlp_options = ["-preload", "tokenize,ssplit,pos,lemma,parse,depparse"]

        jars = list(
            find_jar_iter(
                self._JAR,
                path_to_jar,
                env_vars=("CORENLP",),
                searchpath=(),
                url=_stanford_url,
                verbose=verbose,
                is_regex=True,
            )
        )

        # find the most recent code and model jar
        stanford_jar = max(jars, key=lambda model_name: re.match(self._JAR, model_name))

        if port is None:
            try:
                port = try_port(9000)
            except OSError:
                port = try_port()
                corenlp_options.extend(["-port", str(port)])
        else:
            try_port(port)
            corenlp_options.extend(["-port", str(port)])

        self.url = f"http://localhost:{port}"

        model_jar = max(
            find_jar_iter(
                self._MODEL_JAR_PATTERN,
                path_to_models_jar,
                env_vars=("CORENLP_MODELS",),
                searchpath=(),
                url=_stanford_url,
                verbose=verbose,
                is_regex=True,
            ),
            key=lambda model_name: re.match(self._MODEL_JAR_PATTERN, model_name),
        )

        self.verbose = verbose

        self._classpath = stanford_jar, model_jar

        self.corenlp_options = corenlp_options
        self.java_options = java_options or ["-mx2g"]

    def start(self, stdout="devnull", stderr="devnull"):
        """Starts the CoreNLP server

        :param stdout, stderr: Specifies where CoreNLP output is redirected. Valid values are 'devnull', 'stdout', 'pipe'
        """
        import requests

        cmd = ["edu.stanford.nlp.pipeline.StanfordCoreNLPServer"]

        if self.corenlp_options:
            cmd.extend(self.corenlp_options)

        # Configure java.
        default_options = " ".join(_java_options)
        config_java(options=self.java_options, verbose=self.verbose)

        try:
            self.popen = java(
                cmd,
                classpath=self._classpath,
                blocking=False,
                stdout=stdout,
                stderr=stderr,
            )
        finally:
            # Return java configurations to their default values.
            config_java(options=default_options, verbose=self.verbose)

        # Check that the server is istill running.
        returncode = self.popen.poll()
        if returncode is not None:
            _, stderrdata = self.popen.communicate()
            raise CoreNLPServerError(
                returncode,
                "Could not start the server. "
                "The error was: {}".format(stderrdata.decode("ascii")),
            )

        for i in range(30):
            try:
                response = requests.get(requests.compat.urljoin(self.url, "live"))
            except requests.exceptions.ConnectionError:
                time.sleep(1)
            else:
                if response.ok:
                    break
        else:
            raise CoreNLPServerError("Could not connect to the server.")

        for i in range(60):
            try:
                response = requests.get(requests.compat.urljoin(self.url, "ready"))
            except requests.exceptions.ConnectionError:
                time.sleep(1)
            else:
                if response.ok:
                    break
        else:
            raise CoreNLPServerError("The server is not ready.")

    def stop(self):
        self.popen.terminate()
        self.popen.wait()

    def __enter__(self):
        self.start()

        return self

    def __exit__(self, exc_type, exc_val, exc_tb):
        self.stop()
        return False


class GenericCoreNLPParser(ParserI, TokenizerI, TaggerI):
    """Interface to the CoreNLP Parser."""

    def __init__(
        self,
        url="http://localhost:9000",
        encoding="utf8",
        tagtype=None,
        strict_json=True,
    ):
        import requests

        self.url = url
        self.encoding = encoding

        if tagtype not in ["pos", "ner", None]:
            raise ValueError("tagtype must be either 'pos', 'ner' or None")

        self.tagtype = tagtype
        self.strict_json = strict_json

        self.session = requests.Session()

    def parse_sents(self, sentences, *args, **kwargs):
        """Parse multiple sentences.

        Takes multiple sentences as a list where each sentence is a list of
        words. Each sentence will be automatically tagged with this
        CoreNLPParser instance's tagger.

        If a whitespace exists inside a token, then the token will be treated as
        several tokens.

        :param sentences: Input sentences to parse
        :type sentences: list(list(str))
        :rtype: iter(iter(Tree))
        """
        # Converting list(list(str)) -> list(str)
        sentences = (" ".join(words) for words in sentences)
        return self.raw_parse_sents(sentences, *args, **kwargs)

    def raw_parse(self, sentence, properties=None, *args, **kwargs):
        """Parse a sentence.

        Takes a sentence as a string; before parsing, it will be automatically
        tokenized and tagged by the CoreNLP Parser.

        :param sentence: Input sentence to parse
        :type sentence: str
        :rtype: iter(Tree)
        """
        default_properties = {"tokenize.whitespace": "false"}
        default_properties.update(properties or {})

        return next(
            self.raw_parse_sents(
                [sentence], properties=default_properties, *args, **kwargs
            )
        )

    def api_call(self, data, properties=None, timeout=60):
        default_properties = {
            "outputFormat": "json",
            "annotators": "tokenize,pos,lemma,ssplit,{parser_annotator}".format(
                parser_annotator=self.parser_annotator
            ),
        }

        default_properties.update(properties or {})

        response = self.session.post(
            self.url,
            params={"properties": json.dumps(default_properties)},
            data=data.encode(self.encoding),
            headers={"Content-Type": f"text/plain; charset={self.encoding}"},
            timeout=timeout,
        )

        response.raise_for_status()

        return response.json(strict=self.strict_json)

    def raw_parse_sents(
        self, sentences, verbose=False, properties=None, *args, **kwargs
    ):
        """Parse multiple sentences.

        Takes multiple sentences as a list of strings. Each sentence will be
        automatically tokenized and tagged.

        :param sentences: Input sentences to parse.
        :type sentences: list(str)
        :rtype: iter(iter(Tree))

        """
        default_properties = {
            # Only splits on '\n', never inside the sentence.
            "ssplit.eolonly": "true"
        }

        default_properties.update(properties or {})

        """
        for sentence in sentences:
            parsed_data = self.api_call(sentence, properties=default_properties)

            assert len(parsed_data['sentences']) == 1

            for parse in parsed_data['sentences']:
                tree = self.make_tree(parse)
                yield iter([tree])
        """
        parsed_data = self.api_call("\n".join(sentences), properties=default_properties)
        for parsed_sent in parsed_data["sentences"]:
            tree = self.make_tree(parsed_sent)
            yield iter([tree])

    def parse_text(self, text, *args, **kwargs):
        """Parse a piece of text.

        The text might contain several sentences which will be split by CoreNLP.

        :param str text: text to be split.
        :returns: an iterable of syntactic structures.  # TODO: should it be an iterable of iterables?

        """
        parsed_data = self.api_call(text, *args, **kwargs)

        for parse in parsed_data["sentences"]:
            yield self.make_tree(parse)

    def tokenize(self, text, properties=None):
        """Tokenize a string of text.

        Skip these tests if CoreNLP is likely not ready.
        >>> from nltk.test.setup_fixt import check_jar
        >>> check_jar(CoreNLPServer._JAR, env_vars=("CORENLP",), is_regex=True)

        The CoreNLP server can be started using the following notation, although
        we recommend the `with CoreNLPServer() as server:` context manager notation
        to ensure that the server is always stopped.
        >>> server = CoreNLPServer()
        >>> server.start()
        >>> parser = CoreNLPParser(url=server.url)

        >>> text = 'Good muffins cost $3.88\\nin New York.  Please buy me\\ntwo of them.\\nThanks.'
        >>> list(parser.tokenize(text))
        ['Good', 'muffins', 'cost', '$', '3.88', 'in', 'New', 'York', '.', 'Please', 'buy', 'me', 'two', 'of', 'them', '.', 'Thanks', '.']

        >>> s = "The colour of the wall is blue."
        >>> list(
        ...     parser.tokenize(
        ...         'The colour of the wall is blue.',
        ...             properties={'tokenize.options': 'americanize=true'},
        ...     )
        ... )
        ['The', 'colour', 'of', 'the', 'wall', 'is', 'blue', '.']
        >>> server.stop()

        """
        default_properties = {"annotators": "tokenize,ssplit"}

        default_properties.update(properties or {})

        result = self.api_call(text, properties=default_properties)

        for sentence in result["sentences"]:
            for token in sentence["tokens"]:
                yield token["originalText"] or token["word"]

    def tag_sents(self, sentences):
        """
        Tag multiple sentences.

        Takes multiple sentences as a list where each sentence is a list of
        tokens.

        :param sentences: Input sentences to tag
        :type sentences: list(list(str))
        :rtype: list(list(tuple(str, str))
        """
        # Converting list(list(str)) -> list(str)
        sentences = (" ".join(words) for words in sentences)
        return [sentences[0] for sentences in self.raw_tag_sents(sentences)]

    def tag(self, sentence: str) -> List[Tuple[str, str]]:
        """
        Tag a list of tokens.

        :rtype: list(tuple(str, str))

        Skip these tests if CoreNLP is likely not ready.
        >>> from nltk.test.setup_fixt import check_jar
        >>> check_jar(CoreNLPServer._JAR, env_vars=("CORENLP",), is_regex=True)

        The CoreNLP server can be started using the following notation, although
        we recommend the `with CoreNLPServer() as server:` context manager notation
        to ensure that the server is always stopped.
        >>> server = CoreNLPServer()
        >>> server.start()
        >>> parser = CoreNLPParser(url=server.url, tagtype='ner')
        >>> tokens = 'Rami Eid is studying at Stony Brook University in NY'.split()
        >>> parser.tag(tokens)  # doctest: +NORMALIZE_WHITESPACE
        [('Rami', 'PERSON'), ('Eid', 'PERSON'), ('is', 'O'), ('studying', 'O'), ('at', 'O'), ('Stony', 'ORGANIZATION'),
        ('Brook', 'ORGANIZATION'), ('University', 'ORGANIZATION'), ('in', 'O'), ('NY', 'STATE_OR_PROVINCE')]

        >>> parser = CoreNLPParser(url=server.url, tagtype='pos')
        >>> tokens = "What is the airspeed of an unladen swallow ?".split()
        >>> parser.tag(tokens)  # doctest: +NORMALIZE_WHITESPACE
        [('What', 'WP'), ('is', 'VBZ'), ('the', 'DT'),
        ('airspeed', 'NN'), ('of', 'IN'), ('an', 'DT'),
        ('unladen', 'JJ'), ('swallow', 'VB'), ('?', '.')]
        >>> server.stop()
        """
        return self.tag_sents([sentence])[0]

    def raw_tag_sents(self, sentences):
        """
        Tag multiple sentences.

        Takes multiple sentences as a list where each sentence is a string.

        :param sentences: Input sentences to tag
        :type sentences: list(str)
        :rtype: list(list(list(tuple(str, str)))
        """
        default_properties = {
            "ssplit.isOneSentence": "true",
            "annotators": "tokenize,ssplit,",
        }

        # Supports only 'pos' or 'ner' tags.
        assert self.tagtype in ["pos", "ner"]
        default_properties["annotators"] += self.tagtype
        for sentence in sentences:
            tagged_data = self.api_call(sentence, properties=default_properties)
            yield [
                [
                    (token["word"], token[self.tagtype])
                    for token in tagged_sentence["tokens"]
                ]
                for tagged_sentence in tagged_data["sentences"]
            ]


class CoreNLPParser(GenericCoreNLPParser):
    """
    Skip these tests if CoreNLP is likely not ready.
    >>> from nltk.test.setup_fixt import check_jar
    >>> check_jar(CoreNLPServer._JAR, env_vars=("CORENLP",), is_regex=True)

    The recommended usage of `CoreNLPParser` is using the context manager notation:
    >>> with CoreNLPServer() as server:
    ...     parser = CoreNLPParser(url=server.url)
    ...     next(
    ...         parser.raw_parse('The quick brown fox jumps over the lazy dog.')
    ...     ).pretty_print()  # doctest: +NORMALIZE_WHITESPACE
                            ROOT
                            |
                            S
            _______________|__________________________
            |                         VP               |
            |                _________|___             |
            |               |             PP           |
            |               |     ________|___         |
            NP              |    |            NP       |
        ____|__________     |    |     _______|____    |
        DT   JJ    JJ   NN  VBZ   IN   DT      JJ   NN  .
        |    |     |    |    |    |    |       |    |   |
        The quick brown fox jumps over the     lazy dog  .

    Alternatively, the server can be started using the following notation.
    Note that `CoreNLPServer` does not need to be used if the CoreNLP server is started
    outside of Python.
    >>> server = CoreNLPServer()
    >>> server.start()
    >>> parser = CoreNLPParser(url=server.url)

    >>> (parse_fox, ), (parse_wolf, ) = parser.raw_parse_sents(
    ...     [
    ...         'The quick brown fox jumps over the lazy dog.',
    ...         'The quick grey wolf jumps over the lazy fox.',
    ...     ]
    ... )

    >>> parse_fox.pretty_print()  # doctest: +NORMALIZE_WHITESPACE
                         ROOT
                          |
                          S
           _______________|__________________________
          |                         VP               |
          |                _________|___             |
          |               |             PP           |
          |               |     ________|___         |
          NP              |    |            NP       |
      ____|__________     |    |     _______|____    |
     DT   JJ    JJ   NN  VBZ   IN   DT      JJ   NN  .
     |    |     |    |    |    |    |       |    |   |
    The quick brown fox jumps over the     lazy dog  .

    >>> parse_wolf.pretty_print()  # doctest: +NORMALIZE_WHITESPACE
                         ROOT
                          |
                          S
           _______________|__________________________
          |                         VP               |
          |                _________|___             |
          |               |             PP           |
          |               |     ________|___         |
          NP              |    |            NP       |
      ____|_________      |    |     _______|____    |
     DT   JJ   JJ   NN   VBZ   IN   DT      JJ   NN  .
     |    |    |    |     |    |    |       |    |   |
    The quick grey wolf jumps over the     lazy fox  .

    >>> (parse_dog, ), (parse_friends, ) = parser.parse_sents(
    ...     [
    ...         "I 'm a dog".split(),
    ...         "This is my friends ' cat ( the tabby )".split(),
    ...     ]
    ... )

    >>> parse_dog.pretty_print()  # doctest: +NORMALIZE_WHITESPACE
            ROOT
             |
             S
      _______|____
     |            VP
     |    ________|___
     NP  |            NP
     |   |         ___|___
    PRP VBP       DT      NN
     |   |        |       |
     I   'm       a      dog

    >>> parse_friends.pretty_print()  # doctest: +NORMALIZE_WHITESPACE
         ROOT
          |
          S
      ____|___________
     |                VP
     |     ___________|_____________
     |    |                         NP
     |    |                  _______|________________________
     |    |                 NP           |        |          |
     |    |            _____|_______     |        |          |
     NP   |           NP            |    |        NP         |
     |    |     ______|_________    |    |     ___|____      |
     DT  VBZ  PRP$   NNS       POS  NN -LRB-  DT       NN  -RRB-
     |    |    |      |         |   |    |    |        |     |
    This  is   my  friends      '  cat -LRB- the     tabby -RRB-

    >>> parse_john, parse_mary, = parser.parse_text(
    ...     'John loves Mary. Mary walks.'
    ... )

    >>> parse_john.pretty_print()  # doctest: +NORMALIZE_WHITESPACE
          ROOT
           |
           S
      _____|_____________
     |          VP       |
     |      ____|___     |
     NP    |        NP   |
     |     |        |    |
    NNP   VBZ      NNP   .
     |     |        |    |
    John loves     Mary  .

    >>> parse_mary.pretty_print()  # doctest: +NORMALIZE_WHITESPACE
          ROOT
           |
           S
      _____|____
     NP    VP   |
     |     |    |
    NNP   VBZ   .
     |     |    |
    Mary walks  .

    Special cases

    >>> next(
    ...     parser.raw_parse(
    ...         'NASIRIYA, Iraq—Iraqi doctors who treated former prisoner of war '
    ...         'Jessica Lynch have angrily dismissed claims made in her biography '
    ...         'that she was raped by her Iraqi captors.'
    ...     )
    ... ).height()
    14

    >>> next(
    ...     parser.raw_parse(
    ...         "The broader Standard & Poor's 500 Index <.SPX> was 0.46 points lower, or "
    ...         '0.05 percent, at 997.02.'
    ...     )
    ... ).height()
    11

    >>> server.stop()
    """

    _OUTPUT_FORMAT = "penn"
    parser_annotator = "parse"

    def make_tree(self, result):
        return Tree.fromstring(result["parse"])


class CoreNLPDependencyParser(GenericCoreNLPParser):
    """Dependency parser.

    Skip these tests if CoreNLP is likely not ready.
    >>> from nltk.test.setup_fixt import check_jar
    >>> check_jar(CoreNLPServer._JAR, env_vars=("CORENLP",), is_regex=True)

    The recommended usage of `CoreNLPParser` is using the context manager notation:
    >>> with CoreNLPServer() as server:
    ...     dep_parser = CoreNLPDependencyParser(url=server.url)
    ...     parse, = dep_parser.raw_parse(
    ...         'The quick brown fox jumps over the lazy dog.'
    ...     )
    ...     print(parse.to_conll(4))  # doctest: +NORMALIZE_WHITESPACE
    The        DT      4       det
    quick      JJ      4       amod
    brown      JJ      4       amod
    fox        NN      5       nsubj
    jumps      VBZ     0       ROOT
    over       IN      9       case
    the        DT      9       det
    lazy       JJ      9       amod
    dog        NN      5       obl
    .  .       5       punct

    Alternatively, the server can be started using the following notation.
    Note that `CoreNLPServer` does not need to be used if the CoreNLP server is started
    outside of Python.
    >>> server = CoreNLPServer()
    >>> server.start()
    >>> dep_parser = CoreNLPDependencyParser(url=server.url)
    >>> parse, = dep_parser.raw_parse('The quick brown fox jumps over the lazy dog.')
    >>> print(parse.tree())  # doctest: +NORMALIZE_WHITESPACE
    (jumps (fox The quick brown) (dog over the lazy) .)

    >>> for governor, dep, dependent in parse.triples():
    ...     print(governor, dep, dependent)  # doctest: +NORMALIZE_WHITESPACE
    ('jumps', 'VBZ') nsubj ('fox', 'NN')
    ('fox', 'NN') det ('The', 'DT')
    ('fox', 'NN') amod ('quick', 'JJ')
    ('fox', 'NN') amod ('brown', 'JJ')
    ('jumps', 'VBZ') obl ('dog', 'NN')
    ('dog', 'NN') case ('over', 'IN')
    ('dog', 'NN') det ('the', 'DT')
    ('dog', 'NN') amod ('lazy', 'JJ')
    ('jumps', 'VBZ') punct ('.', '.')

    >>> (parse_fox, ), (parse_dog, ) = dep_parser.raw_parse_sents(
    ...     [
    ...         'The quick brown fox jumps over the lazy dog.',
    ...         'The quick grey wolf jumps over the lazy fox.',
    ...     ]
    ... )
    >>> print(parse_fox.to_conll(4))  # doctest: +NORMALIZE_WHITESPACE
    The        DT      4       det
    quick      JJ      4       amod
    brown      JJ      4       amod
    fox        NN      5       nsubj
    jumps      VBZ     0       ROOT
    over       IN      9       case
    the        DT      9       det
    lazy       JJ      9       amod
    dog        NN      5       obl
    .  .       5       punct

    >>> print(parse_dog.to_conll(4))  # doctest: +NORMALIZE_WHITESPACE
    The        DT      4       det
    quick      JJ      4       amod
    grey       JJ      4       amod
    wolf       NN      5       nsubj
    jumps      VBZ     0       ROOT
    over       IN      9       case
    the        DT      9       det
    lazy       JJ      9       amod
    fox        NN      5       obl
    .  .       5       punct

    >>> (parse_dog, ), (parse_friends, ) = dep_parser.parse_sents(
    ...     [
    ...         "I 'm a dog".split(),
    ...         "This is my friends ' cat ( the tabby )".split(),
    ...     ]
    ... )
    >>> print(parse_dog.to_conll(4))  # doctest: +NORMALIZE_WHITESPACE
    I   PRP     4       nsubj
    'm  VBP     4       cop
    a   DT      4       det
    dog NN      0       ROOT

    >>> print(parse_friends.to_conll(4))  # doctest: +NORMALIZE_WHITESPACE
    This       DT      6       nsubj
    is VBZ     6       cop
    my PRP$    4       nmod:poss
    friends    NNS     6       nmod:poss
    '  POS     4       case
    cat        NN      0       ROOT
    (  -LRB-   9       punct
    the        DT      9       det
    tabby      NN      6       dep
    )  -RRB-   9       punct

    >>> parse_john, parse_mary, = dep_parser.parse_text(
    ...     'John loves Mary. Mary walks.'
    ... )

    >>> print(parse_john.to_conll(4))  # doctest: +NORMALIZE_WHITESPACE
    John       NNP     2       nsubj
    loves      VBZ     0       ROOT
    Mary       NNP     2       obj
    .  .       2       punct

    >>> print(parse_mary.to_conll(4))  # doctest: +NORMALIZE_WHITESPACE
    Mary        NNP     2       nsubj
    walks       VBZ     0       ROOT
    .   .       2       punct

    Special cases

    Non-breaking space inside of a token.

    >>> len(
    ...     next(
    ...         dep_parser.raw_parse(
    ...             'Anhalt said children typically treat a 20-ounce soda bottle as one '
    ...             'serving, while it actually contains 2 1/2 servings.'
    ...         )
    ...     ).nodes
    ... )
    23

    Phone  numbers.

    >>> len(
    ...     next(
    ...         dep_parser.raw_parse('This is not going to crash: 01 111 555.')
    ...     ).nodes
    ... )
    10

    >>> print(
    ...     next(
    ...         dep_parser.raw_parse('The underscore _ should not simply disappear.')
    ...     ).to_conll(4)
    ... )  # doctest: +NORMALIZE_WHITESPACE
    The        DT      2       det
    underscore NN      7       nsubj
    _  NFP     7       punct
    should     MD      7       aux
    not        RB      7       advmod
    simply     RB      7       advmod
    disappear  VB      0       ROOT
    .  .       7       punct

    >>> print(
    ...     next(
    ...         dep_parser.raw_parse(
    ...             'for all of its insights into the dream world of teen life , and its electronic expression through '
    ...             'cyber culture , the film gives no quarter to anyone seeking to pull a cohesive story out of its 2 '
    ...             '1/2-hour running time .'
    ...         )
    ...     ).to_conll(4)
    ... )  # doctest: +NORMALIZE_WHITESPACE +ELLIPSIS
    for        IN      2       case
    all        DT      24      obl
    of IN      5       case
    its        PRP$    5       nmod:poss
    insights   NNS     2       nmod
    into       IN      9       case
    the        DT      9       det
    dream      NN      9       compound
    world      NN      5       nmod
    of IN      12      case
    teen       NN      12      compound
    ...

    >>> server.stop()
    """

    _OUTPUT_FORMAT = "conll2007"
    parser_annotator = "depparse"

    def make_tree(self, result):

        return DependencyGraph(
            (
                " ".join(n_items[1:])  # NLTK expects an iterable of strings...
                for n_items in sorted(transform(result))
            ),
            cell_separator=" ",  # To make sure that a non-breaking space is kept inside of a token.
        )


def transform(sentence):
    for dependency in sentence["basicDependencies"]:

        dependent_index = dependency["dependent"]
        token = sentence["tokens"][dependent_index - 1]

        # Return values that we don't know as '_'. Also, consider tag and ctag
        # to be equal.
        yield (
            dependent_index,
            "_",
            token["word"],
            token["lemma"],
            token["pos"],
            token["pos"],
            "_",
            str(dependency["governor"]),
            dependency["dep"],
            "_",
            "_",
        )