Source code for gatenlp.stanfordnlp
"""
Support for using stanfordnlp: convert from stanfordnlp output to gatenlp documents and annotations.
"""
from gatenlp import Document
from gatenlp import utils
[docs]def apply_stanfordnlp(nlp, gatenlpdoc, setname=""):
"""
Run the stanford nlp pipeline on the gatenlp document and transfer the annotations.
This modifies the gatenlp document in place.
:param nlp: StanfordNLP pipeline
:param gatenlpdoc: gatenlp document
:param setname: set to use
:return:
"""
spacydoc = nlp(gatenlpdoc.text)
return stanfordnlp2gatenlp(spacydoc, gatenlpdoc=gatenlpdoc, setname=setname)
[docs]def stanfordnlp2gatenlp(stanfordnlpdoc, gatenlpdoc=None, setname="", word_type="Word",
sentence_type="Sentence"):
"""
Convert a StanfordNLP document to a gatenlp document. If a gatenlp document is already
provided, add the annotations from the spacy document to it. In this case the
original gatenlpdoc is used and gets modified.
:param stanfordnlpdoc: a spacy document
:param gatenlpdoc: if None, a new gatenlp document is created otherwise this
document is added to.
:param setname: the annotation set name to which the annotations get added, empty string
for the default annotation set.
:param token_type: the annotation type to use for tokens
:param sentence_type: the annotation type to use for sentence anntoations
:return: the new or modified
"""
if gatenlpdoc is None:
retdoc = Document(stanfordnlpdoc.text)
else:
retdoc = gatenlpdoc
toki2annid = {}
annset = retdoc.get_annotations(setname)
# stanford nlp processes text in sentence chunks, so we do everything per sentence
# NOTE: the stanford elements do not contain any text offsets, so we have to match and find
# them ourselves. for this we keep an index to first character in the text which has not
# been matched yet
notmatchedidx = 0
for sent in stanfordnlpdoc.sentences:
# a sentence is a list of tokens and a list of words. Some tokens consist of several words.
# dependency parsers are over words, so we create Word and Token annotations, but we only
# set the features per Word annotation for now.
offsetinfos = utils.match_substrings(stanfordnlpdoc.text[notmatchedidx:],
sent.words, getstr=lambda x: x.text)
idx2annid = {}
for oinfo in offsetinfos:
word = oinfo[2]
fm = {
"string": word.text,
"lemma": word.lemma,
"upos": word.upos,
"xpos": word.xpos,
"dependency_relation": word.dependency_relation,
"governor": int(word.governor)
}
for feat in word.feats.split("|"):
if feat and feat != "_":
k, v = feat.split("=")
# TODO: maybe try to detect and convert bool/int values
fm["feat_"+k] = v
snlp_idx = int(word.index)
annid = annset.add(oinfo[0]+notmatchedidx, oinfo[1]+notmatchedidx, word_type, fm)
idx2annid[snlp_idx] = annid
# create a sentence annotation from beginning of first word to end of last
sentid = annset.add(offsetinfos[0][0]+notmatchedidx, offsetinfos[-1][1]+notmatchedidx, sentence_type)
# now replace the governor index with the corresponding annid, the governor index is
# mapped to the sentence annotation
idx2annid[0] = sentid
for annid in list(idx2annid.values()):
ann = annset.get(annid)
gov = ann.get_feature("governor")
if gov is not None:
ann.set_feature("governor", idx2annid[gov])
notmatchedidx = offsetinfos[-1][1]+notmatchedidx + 1
return retdoc