class StanfordParser(object):
def __init__(self, nlp=None, annots=None, props=None):
if annots is None:
annots = "tokenize pos lemma depparse"
if nlp is None:
self.nlp_client = CoreNLPClient(annotators=annots,
output_format='json')
else:
self.nlp_client = nlp
if props is not None:
self.nlp_client.default_properties.update(props)
_ = self.nlp_client.annotate("Let's get this party started!")
del (_)
def get_parse(self, sentence):
return self.nlp_client.annotate(sentence)
def get_deps(self, sentence, deptype='basicDependencies', ret='asis'):
if isinstance(sentence, str):
sentence = self.get_parse(sentence)['sentences'][0]
deps = sentence[deptype]
if ret == 'asis':
retval = deps
else:
retval = {}
retval['deps'] = {x['dep']: x['dependent'] for x in deps}
retval['heads'] = {
x['dependentGloss']: x['governorGloss']
for x in deps
}
retval['governors'] = {
x['dependent']: x['governorGloss']
for x in deps
}
retval['dependents'] = {
x['dependent']: x['dependentGloss']
for x in deps
}
retval['text'] = [
"{}({}-{}, {}-{})".format(x['dep'], x['governorGloss'],
x['governor'], x['dependentGloss'],
x['dependent']) for x in deps
]
return retval
class StanfordService:
def __init__(self, parser_path: str):
os.environ['JAVANLP_HOME'] = parser_path
print(
'starting CoreNLP server with JAVANLP_HOME {}'.format(parser_path))
self.nlp = CoreNLPClient(annotators="tokenize ssplit".split(),
timeout=1000000)
def tokenize(self, text: str) -> List[Token]:
for _ in range(10):
try:
annotated_result = self.nlp.annotate(text)
stanford_document = StanfordDocument.from_proto(
annotated_result)
return StanfordService.idiomatic_tokens(stanford_document)
except:
print('exception while annotating result')
sleep(10)
@staticmethod
def idiomatic_tokens(doc: StanfordDocument):
stanford_tokens = [
token for sentence in doc.sentences for token in sentence.tokens
]
return [
StanfordService.idiomatic_token(token, index)
for index, token in enumerate(stanford_tokens)
]
@staticmethod
def idiomatic_token(token: StanfordToken, token_index: int) -> Token:
return Token(token.originalText, token_index,
token.characterOffsetBegin)
class StanfordCoreferenceResolution:
"""
Stanford CoreNLP co-reference.
Parameters
----------
timeout : int
The timeout for the parser
Defaults to 30000
memory : str
The memory allocation.
Defaults to '6G'
"""
def __init__(self, timeout=30000, memory='6G'):
self.detok = TreebankWordDetokenizer()
self.client = CoreNLPClient(
annotators=['tokenize', 'ssplit', 'dcoref'],
output_format='json',
timeout=timeout,
memory=memory)
def resolve(self, doc, raise_errors=True):
"""
Resolve the co-references for a single document.
Parameters
----------
doc : str
A document whose co-references will be resolved.
raise_errors : bool, optional
Whether to raise errors.
Defaults to True.
Returns
-------
resolve_doc : str or None
A document whose co-references have been resolved.
If there was a problem and `raise_errors=False`,
then `None` will be returned.
"""
try:
parsed = self.client.annotate(doc)
except Exception as error:
if raise_errors:
raise error
return
return self.replace_coreferences(parsed)
def resolve_all(self, docs, raise_errors=True):
"""
Resolve co-references for all the documents.
Parameters
----------
docs : list of str
A list of documents
raise_errors : bool, optional
Whether to raise errors.
Defaults to False.
Returns
-------
resolved_docs : list of str
A list of documents, with co-references resolved.
"""
resolved_docs = []
for doc in tqdm(docs):
resolved_docs.append(self.resolve(doc, raise_errors))
return resolved_docs
@staticmethod
def restructure_coreference_dict(corefs_dict):
"""
Given a dictionary of co-references, restructure it into
a new dictionary where the keys are sentence numbers
and the values are lists of references that need to
be resolved.
Parameters
----------
corefs_dict : dict
A co-reference dictionary, output from Stanford.
"""
corefs_list = [
corefs_dict[key] for key in corefs_dict
if len(corefs_dict[key]) > 1 and any(
not co['isRepresentativeMention'] for co in corefs_dict[key])
]
corefs_dict = defaultdict(list)
for i, coref in enumerate(corefs_list):
# get the first representative mention from the list;
# if there are no representative mentions, continue
represent = [
co['text'] for co in coref if co['isRepresentativeMention']
]
if len(represent) >= 1:
represent = represent[0]
else:
continue
# loop through the (non-representative) mentions,
# add to the dictionary list for that sentence
for co in coref:
if not co['isRepresentativeMention']:
mention = {
'represent': represent,
'text': co['text'],
'startIndex': co['startIndex'],
'endIndex': co['endIndex'],
'sentNum': co['sentNum']
}
corefs_dict[co['sentNum']].append(mention)
return corefs_dict
def replace_coreferences(self, parsed):
"""
We want to replace all the references with their
representative mention.
Parameters
----------
parsed : dict
The full output from Stanford, with co-references and sentences.
"""
corefs = parsed['corefs']
sents = parsed['sentences']
corefs_dict = self.restructure_coreference_dict(corefs)
sents = [[s['word'] for s in sent['tokens']] for sent in sents]
sents_new = []
# we do this on a sentence-by-sentence basis
for sent_i, sent in enumerate(sents, start=1):
sent_new = []
# we check to see if the sentence is in the co-reference dictionary;
# if it's not we won't need to do anything.
if sent_i in corefs_dict:
last_end = 0
# we loop through the (sorted) references and add them
# to our new sentence list one-by-one, being careful to
# capture any preceding or ending text
sorted_sent = sorted(corefs_dict[sent_i],
key=lambda x: x['startIndex'])
for co_i, co in enumerate(sorted_sent):
start = co['startIndex'] - 1
end = co['endIndex'] - 1
represent = co['represent']
# here we want to check whether this is the first co-reference;
# if it is, then we need to get any text *before* it
if co_i == 0:
sent_new.extend(sent[:start])
sent_new.append(represent[0].upper() +
represent[1:] if start ==
0 else represent)
# otherwise, we just get the co-reference and anything
# between it and the preceding end from the previous co-reference
else:
sent_new.extend(sent[last_end:start])
sent_new.append(represent)
last_end = end
sent_new.extend(sent[last_end:])
else:
sent_new = sent
sents_new.append(sent_new)
# we need to detokenize the sentence; basically this handles
# putting punctuation and weird symbols for parentheses back together
sents = ' '.join([
self.detok.detokenize(sent, convert_parentheses=True)
for sent in sents_new
])
return sents
def prepro_each(args,
data_type,
start_ratio=0.0,
stop_ratio=1.0,
out_name="default",
in_path=None):
if args.tokenizer == "PTB":
import nltk.tokenize as nltk
sent_tokenize = nltk.sent_tokenize
def word_tokenize(tokens):
return [
token.replace("''", '"').replace("``", '"')
for token in nltk.word_tokenize(tokens)
]
elif args.tokenizer == 'Stanford':
from corenlp import CoreNLPClient
interface = CoreNLPClient(annotators="tokenize ssplit".split())
else:
raise Exception()
"""
if not args.split:
sent_tokenize = lambda para: [para]
"""
source_path = in_path or os.path.join(
args.source_dir, "{}-{}v1.1.json".format(data_type, args.suffix))
source_data = json.load(open(source_path, 'r'))
q, cq, y, rx, rcx, ids, idxs = [], [], [], [], [], [], []
na = []
cy = []
x, cx = [], []
answerss = []
p = []
word_counter, char_counter, lower_word_counter = Counter(), Counter(
), Counter()
start_ai = int(round(len(source_data['data']) * start_ratio))
stop_ai = int(round(len(source_data['data']) * stop_ratio))
for ai, article in enumerate(tqdm(source_data['data'][start_ai:stop_ai])):
xp, cxp = [], []
pp = []
x.append(xp)
cx.append(cxp)
p.append(pp)
for pi, para in enumerate(article['paragraphs']):
# wordss
context = para['context']
context = context.replace("''", '" ')
context = context.replace("``", '" ')
while True:
try:
temp = interface.annotate(context)
break
except Exception as e:
time.sleep(0.2)
context_s = []
for sent in temp.sentence:
sent = [word.originalText for word in sent.token]
# CoreNLP 가 처리못하는 단어인 \xa0 를 예외처리하기위한 수동 방법...
for wi in range(len(sent)):
if "\xa0" in sent[wi]:
sent = sent[:wi] + sent[wi].split("\xa0") + sent[wi +
1:]
wi = 0
context_s.append(sent)
xi = context_s
xi = [process_tokens(tokens) for tokens in xi] # process tokens
# given xi, add chars
cxi = [[list(xijk) for xijk in xij] for xij in xi]
xp.append(xi)
cxp.append(cxi)
pp.append(context)
for xij in xi:
for xijk in xij:
word_counter[xijk] += len(para['qas'])
lower_word_counter[xijk.lower()] += len(para['qas'])
for xijkl in xijk:
char_counter[xijkl] += len(para['qas'])
rxi = [ai, pi]
assert len(x) - 1 == ai
assert len(x[ai]) - 1 == pi
for qa in para['qas']:
# get words
while True:
try:
temp = interface.annotate(qa['question']).sentence[0]
break
except Exception as e:
time.sleep(0.2)
#print(temp.token[0])
#exit(-1)
qi = [t_s.originalText for t_s in temp.token]
print(qi)
exit(-1)
qi = process_tokens(qi)
cqi = [list(qij) for qij in qi]
yi = []
cyi = []
answers = []
for answer in qa['answers']:
answer_text = answer['text']
answers.append(answer_text)
answer_start = answer['answer_start']
answer_stop = answer_start + len(answer_text)
# TODO : put some function that gives word_start, word_stop here
yi0, yi1 = get_word_span(context, xi, answer_start,
answer_stop)
# yi0 = answer['answer_word_start'] or [0, 0]
# yi1 = answer['answer_word_stop'] or [0, 1]
assert len(xi[yi0[0]]) > yi0[1]
assert len(xi[yi1[0]]) >= yi1[1]
w0 = xi[yi0[0]][yi0[1]]
w1 = xi[yi1[0]][yi1[1] - 1]
i0 = get_word_idx(context, xi, yi0)
i1 = get_word_idx(context, xi, (yi1[0], yi1[1] - 1))
cyi0 = answer_start - i0
cyi1 = answer_stop - i1 - 1
# print(answer_text, w0[cyi0:], w1[:cyi1+1])
assert answer_text[0] == w0[cyi0], (answer_text, w0, cyi0)
assert answer_text[-1] == w1[cyi1]
assert cyi0 < 32, (answer_text, w0)
assert cyi1 < 32, (answer_text, w1)
yi.append([yi0, yi1])
cyi.append([cyi0, cyi1])
if len(qa['answers']) == 0:
yi.append([(0, 0), (0, 1)])
cyi.append([0, 1])
na.append(True)
else:
na.append(False)
for qij in qi:
word_counter[qij] += 1
lower_word_counter[qij.lower()] += 1
for qijk in qij:
char_counter[qijk] += 1
q.append(qi)
cq.append(cqi)
y.append(yi)
cy.append(cyi)
rx.append(rxi)
rcx.append(rxi)
ids.append(qa['id'])
idxs.append(len(idxs))
answerss.append(answers)
if args.debug:
break
word2vec_dict = get_word2vec(args, word_counter)
lower_word2vec_dict = get_word2vec(args, lower_word_counter)
# add context here
data = {
'q': q,
'cq': cq,
'y': y,
'*x': rx,
'*cx': rcx,
'cy': cy,
'idxs': idxs,
'ids': ids,
'answerss': answerss,
'*p': rx,
'na': na
}
shared = {
'x': x,
'cx': cx,
'p': p,
'word_counter': word_counter,
'char_counter': char_counter,
'lower_word_counter': lower_word_counter,
'word2vec': word2vec_dict,
'lower_word2vec': lower_word2vec_dict
}
print("saving ...")
save(args, data, shared, out_name)
class Featurizer():
def __init__(self, annotators=None, properties=None):
with open(
os.path.join(os.path.dirname(__file__), "assets",
"regexes.json")) as f:
self.regexer = RegexFeaturizer(json.load(f))
self.annotators = annotators or CORENLP_ANNOTATORS
self.properties = properties or CORENLP_PROPERTIES
self.client = CoreNLPClient(self.annotators,
properties=self.properties,
endpoint="http://localhost:9012")
def __enter__(self):
self.client.__enter__()
return self
def __exit__(self, *args):
self.client.__exit__(*args)
def _apply_features(self, obj, ann=None):
"""
Adds features to a graph.
"""
if "features" not in obj:
obj["features"] = {}
if ann:
assert len(ann.sentence) == 1
sentence = ann.sentence[0]
assert len(sentence.token) == len(obj["tokens"])
obj["features"]["lemma"] = [t.lemma for t in sentence.token]
obj["features"]["pos"] = [t.pos for t in sentence.token]
obj["features"]["ner"] = [t.ner for t in sentence.token]
obj["features"]["depparse"] = _dep_to_list(
sentence.enhancedPlusPlusDependencies)
assert len({tail
for _, tail, _ in obj["features"]["depparse"]
}) == len(sentence.token)
child_to_head, head_to_child, path_length, next_in_path, distance_to_next_token, distance_from_prev_token = compute_dependency_paths(
obj)
obj["features"]["dep_child_to_head"] = child_to_head
obj["features"]["dep_head_to_child"] = head_to_child
obj["features"]["dep_path_lengths"] = path_length
obj["features"]["dep_traceback"] = next_in_path
obj["features"]["dep_dist_to_next"] = distance_to_next_token
obj["features"]["dep_dist_from_prev"] = distance_from_prev_token
if self.regexer:
obj["features"]["regexes"] = self.regexer.featurize(obj["tokens"])
obj["features"][
"typed_values"] = self.regexer.featurize_unit_spans(
obj["tokens"])
def featurize_graph(self, obj):
ann = self.client.annotate(" ".join(obj["tokens"]))
self._apply_features(obj, ann)
return obj
def featurize_text(self, text):
ann = self.client.annotate(text)
assert len(ann.sentence) == 1
sentence = ann.sentence[0]
obj = {
"tokens": [t.word for t in sentence.token],
}
self._apply_features(obj, ann)
return obj