def handle(self, *args, **options):
spacy_model = spacy.load(settings.SPACY_CUSTOMN_MODEL_FOLDER, disable=['parser', 'tagger', 'textcat'])
Span.set_extension('is_phone', getter=Command.is_phone_getter, force=True)
Span.set_extension('line_number', getter=Command.line_number_getter, force=True)
Doc.set_extension('lines', getter=Command.get_lines, setter=Command.set_lines)
Doc.set_extension('_lines', default=list())
logger.debug("Loaded spacy server")
main_socks, read_socks, write_socks = socket_bind('', settings.SPACY_PORT)
while True:
readable, writeable, exceptions = select(read_socks, write_socks, [])
for sockobj in readable:
if sockobj in main_socks:
new_sock, address = sockobj.accept()
logger.debug('Connect: %s - %s', address, id(new_sock))
read_socks.append(new_sock)
else:
try:
entities = []
data = recv_end(sockobj)
if not data:
sockobj.close()
read_socks.remove(sockobj)
else:
for doc in spacy_model.pipe([data]):
doc._.lines = [x.start() for x in re.finditer('\n', doc.text)]
for ent in doc.ents:
current_entity = self.get_ent(ent)
entities.append(current_entity) if current_entity else None
sockobj.sendall(json.dumps(entities).encode('utf8') + '--end--'.encode('utf8'))
except:
pass
def set_attributes(self, schema_file: Union[str, Path] = '', encoding: str = None) -> Set:
"""
The current version SpaCy doesn't differentiate attributes for different annotation types.
Thus, any attributes extended here will be applied to all Spans.
@param schema_file: initiate Span attributes using eHOST schema configuration file
@param encoding: text encoding
@return: a set of attribute names
"""
schema_file = self.check_file_validity(schema_file, False)
attr_names = set()
attr_conf_start = False
if schema_file is not None and schema_file.name.endswith("conf"):
for row in schema_file.read_text(encoding=encoding).split("\n"):
if len(row.strip()) == 0 or row[0] == '#':
continue
if row.startswith(r'[attributes]'):
attr_conf_start = True
continue
elif row[0] == '[':
attr_conf_start = False
if attr_conf_start:
# [attributes]
# Negation Arg:<EVENT>
# Confidence Arg:<EVENT>, Value:Possible|Likely|Certain
name = row.split(' ')[0]
default_value = None
if name not in attr_names and not Span.has_extension(name):
Span.set_extension(name, default=default_value)
attr_names.add(name)
self.schema_set = True
return attr_names
def __init__(self, nlp: Language = None, support_overlap: bool = False,
log_level: int = logging.WARNING, encoding: str = None, doc_name_depth: int = 0,
schema_file: Union[str, Path] = '', store_anno_string: bool = False,
**kwargs):
"""
@param nlp: Spacy Language model
@param support_overlap: whether need to support overlapped annotations
@param log_level: logging level configuration
@param encoding: txt encoding
@param doc_name_depth: depth of parent directories to add into doc_name
default is 0: only use file name
1: use 1 level parent directory name + file name
-1: use full absolution path
if you are dealing with multiple directories,this is helpful to
locate the original files
@param schema_file: initiate Span attributes using eHOST schema configuration file
@param store_anno_string: whether read annotated string from annotations to double check parsed Span's correction
@param kwargs:other parameters
"""
self.schema_set = False
self.attr_names = self.set_attributes(schema_file=schema_file, encoding=encoding)
if store_anno_string:
if not Span.has_extension("span_txt"):
Span.set_extension("span_txt", default="")
super().__init__(nlp=nlp, support_overlap=support_overlap,
log_level=log_level, encoding=encoding, doc_name_depth=doc_name_depth,
schema_file=schema_file, store_anno_string=store_anno_string, **kwargs)
pass
def add_span_extensions():
Doc.set_extension("relations", default=None)
Doc.set_extension("entities", default=None)
for span_extension in [
'entity_type', 'entity_id', 'foodon', 'hansard', 'hansardClosest',
'hansardParent', 'snomedct', 'synonyms'
]:
Span.set_extension(span_extension, default=None)
def one_verb(i, sent, doc):
front = ""
if (i > 0):
front = Span(doc, sent.start, i+sent.start)
end = Span(doc, sent.start+i+1, sent.end-1)
newfront = ""
for k in front:
if k.ent_type_ == "":
newfront += k.lower_ + " "
else:
newfront += k.text + " "
quest1 = ""
quest2 = ""
if sent[i].tag_ == "VBD" and sent[i].lemma_ != "be":
if find_subject(sent) != "UNKNOWN":
quest1 = "did"+ " " + find_subject(sent) + " " + sent[i].lemma_
quest2 = "did"+ " " + str(newfront[:-1]) + " " + sent[i].lemma_
elif sent[i].tag_ == "VBZ" and sent[i].lemma_ != "be":
if find_subject(sent) != "UNKNOWN":
quest1 = "does"+ " " + find_subject(sent) + " " + sent[i].lemma_
quest2 = "does"+ " " + str(newfront[:-1]) + " " + sent[i].lemma_
elif sent[i].tag_ == "VBP" or sent[i].tag_ == "VB":
if find_subject(sent) != "UNKNOWN":
quest1 = "do"+ " " + find_subject(sent) + " " + sent[i].lemma_
quest2 = "do"+ " " + str(newfront[:-1]) + " " + sent[i].lemma_
elif sent[i].tag_ == "VBZ" and sent[i].lemma_ == "be":
if find_subject(sent) != "UNKNOWN":
quest1 = sent[i].orth_ + " " + find_subject(sent)
quest2 = sent[i].orth_ + " " + str(newfront[:-1])
elif sent[i].tag_ == "VBD" and sent[i].lemma_ == "be":
if find_subject(sent) != "UNKNOWN":
quest1 = "was"+ " " + find_subject(sent)
quest2 = "was"+ " " + str(newfront[:-1])
elif sent[i].tag_ == "VBN" and sent[i].lemma_ == "be":
if find_subject(sent) != "UNKNOWN":
quest1 = sent[i].orth_ + " " + find_subject(sent) + " " + sent[i].lemma_
quest2 = sent[i].orth_ + " " + str(newfront[:-1]) + " " + sent[i].lemma_
elif sent[i].tag_ == "VBN" and sent[i].lemma_ == "have":
if find_subject(sent) != "UNKNOWN":
quest1 = sent[i].orth_ + " " + find_subject(sent) + " " + sent[i].lemma_
quest2 = sent[i].orth_ + " " + str(newfront[:-1]) + " " + sent[i].lemma_
quest1 += " " + str(end) + "?"
quest2 += " " + str(end) + "?"
return quest1, quest2
def __init__(self, language: str = "es"):
"""
Init method
:param language: language of the annotation
"""
self.__sentiment_words = load_dict(language, "sentiment_words.csv")
self.__boosters = load_dict(language, "boosters.csv")
self.__negations = load_dict(language, "negations.csv")
Span.set_extension("sentiment_weight", default=0.0, force=True)
Token.set_extension("sentiment_weight", default=0.0, force=True)
Token.set_extension("negation_weight", default=1.0, force=True)
Token.set_extension("booster_weight", default=0.0, force=True)
def test_ensure_print_span_characteristics_wont_fail():
"""Test if interface between two methods aren't destroyed if refactored"""
nlp = English()
spans_key = "sc"
pred = Doc(nlp.vocab, words=["Welcome", "to", "the", "Bank", "of", "China", "."])
pred.spans[spans_key] = [Span(pred, 3, 6, "ORG"), Span(pred, 5, 6, "GPE")]
ref = Doc(nlp.vocab, words=["Welcome", "to", "the", "Bank", "of", "China", "."])
ref.spans[spans_key] = [Span(ref, 3, 6, "ORG"), Span(ref, 5, 6, "GPE")]
eg = Example(pred, ref)
examples = [eg]
data = _compile_gold(examples, ["spancat"], nlp, True)
span_characteristics = _get_span_characteristics(
examples=examples, compiled_gold=data, spans_key=spans_key
)
_print_span_characteristics(span_characteristics)
def __init__(self,
nlp: Language = None,
support_overlap: bool = False,
log_level: int = logging.WARNING,
encoding: str = None,
doc_name_depth: int = 0,
schema_file: Union[str, Path] = '',
store_anno_string: bool = False,
use_adjudication: bool = False,
**kwargs):
"""
@param nlp: a SpaCy language model
@param support_overlap: if the EhostDocReader need to support reading from overlapped annotations.
Because SpaCy's Doc.ents does not allows overlapped Spans, to support overlapping, Spans need to be stored
somewhere else----Doc._.concepts
@param log_level: set the logger's logging level. TO debug, set to logging.DEBUG
@param encoding: txt encoding
@param doc_name_depth: depth of parent directories to add into doc_name
default is 0: only use file name
1: use 1 level parent directory name + file name
-1: use full absolution path
if you are dealing with multiple directories,this is helpful to
locate the original files
@param schema_file: initiate Span attributes using eHOST schema configuration file
@param store_anno_string: whether read annotated string from annotations to double check parsed Span's correction
@param use_adjudication: if read annotations from adjudication folder
@param kwargs:other parameters
"""
self.schema_set = False
self.attr_names = self.set_attributes(schema_file=schema_file,
encoding=encoding)
if store_anno_string:
if not Span.has_extension("span_txt"):
Span.set_extension("span_txt", default="")
super().__init__(nlp=nlp,
support_overlap=support_overlap,
log_level=log_level,
encoding=encoding,
doc_name_depth=doc_name_depth,
schema_file=schema_file,
store_anno_string=store_anno_string,
use_adjudication=use_adjudication,
**kwargs)
pass
def __call__(self, spacy_span: Span, describer=None):
"""
convenient wrapper around make_issue if you are using spaCy
usage example:
```python
from spacy.tokens import Span
from app.factor import SpacyFactor
SOV = SpacyFactor(
"subject_object_verb_spacing",
"Keep the subject, verb, and object of a sentence close together to help the reader understand the sentence."
)
Span.set_extension("score", default=0)
Span.set_extension("suggestions", default=[])
doc = nlp("Holders of the Class A and Class B-1 certificates will be entitled to receive on each Payment Date, to the extent monies are available therefor (but not more than the Class A Certificate Balance or Class B-1 Certificate Balance then outstanding), a distribution.")
score = analyze(doc)
if score is not None:
span = Span(doc, 0, len(doc)) # or whichever TOKENS are the issue (don't have to worry about character indexes)
span._.score = score
span._.suggestions = get_suggestions(doc)
issues = SOV(span)
```
"""
text, start, end = spacy_span.text, spacy_span.start_char, spacy_span.end_char
score = spacy_span._.score if spacy_span.has_extension("score") else 0
suggestions = (spacy_span._.suggestions
if spacy_span.has_extension("suggestions") else [])
if describer:
description = describer(spacy_span)
else:
description = self.description
return make_issue(
text,
start,
end,
issue_type=self.issue_type,
score=score,
description=description,
suggestions=suggestions,
)
def __init__(self,
first_name_extension_name=FirstNameListMatcher.EXTENSION_NAME,
last_name_extension_name=LastNameListMatcher.EXTENSION_NAME):
self.token_extension_name = self.TOKEN_EXTENSION_NAME
self.span_extension_name = self.SPAN_EXTENSION_NAME
self.doc_extension_name = self.DOC_EXTENSION_NAME
self.first_name_extension_name = first_name_extension_name
self.last_name_extension_name = last_name_extension_name
if not Token.has_extension(self.token_extension_name):
Token.set_extension(self.token_extension_name,
default=self.ANOT_NONE)
if not Span.has_extension(self.span_extension_name):
Span.set_extension(self.span_extension_name,
getter=self.is_full_name_getter)
if not Doc.has_extension(self.doc_extension_name):
Doc.set_extension(self.doc_extension_name, default=[])
def to_and_from(doc):
new_ents = []
for ent in doc.ents:
# Only check for title if it's a station and not the first token
if ent.label_ == "STN" and ent.start != 0:
prev_token = doc[ent.start - 1]
if prev_token.text.lower() in ("to", "at"):
new_ent = Span(doc, ent.start, ent.end, label="ARR")
new_ents.append(new_ent)
elif prev_token.text.lower() in "from":
new_ent = Span(doc, ent.start, ent.end, label="DEP")
new_ents.append(new_ent)
else:
new_ents.append(ent)
else:
new_ents.append(ent)
doc.ents = new_ents
return doc
def test_get_span_characteristics_return_value():
nlp = English()
spans_key = "sc"
pred = Doc(nlp.vocab, words=["Welcome", "to", "the", "Bank", "of", "China", "."])
pred.spans[spans_key] = [Span(pred, 3, 6, "ORG"), Span(pred, 5, 6, "GPE")]
ref = Doc(nlp.vocab, words=["Welcome", "to", "the", "Bank", "of", "China", "."])
ref.spans[spans_key] = [Span(ref, 3, 6, "ORG"), Span(ref, 5, 6, "GPE")]
eg = Example(pred, ref)
examples = [eg]
data = _compile_gold(examples, ["spancat"], nlp, True)
span_characteristics = _get_span_characteristics(
examples=examples, compiled_gold=data, spans_key=spans_key
)
assert {"sd", "bd", "lengths"}.issubset(span_characteristics.keys())
assert span_characteristics["min_length"] == 1
assert span_characteristics["max_length"] == 3
def add_new_matching_span(self, document, span):
token_end_index, token_start_index = self.calculate_span_indices(document, span)
length = span[1] - span[0]
if self.config.minimum_length is not None and self.config.minimum_length > length:
return
if self.config.maximum_length is not None and self.config.maximum_length < length:
return
spacy_span = Span(document, token_start_index, token_end_index, self.config.label)
document.ents = document.ents + (spacy_span,)
def __init__(self, links, **kwargs):
self.start_urls.append(links)
import spacy
from spacy.tokens.doc import Doc
from spacy.tokens.span import Span
self.spacy_model = spacy.load(settings.SPACY_CUSTOMN_MODEL_FOLDER)
Span.set_extension('line_number',
getter=TagLinkSpider.line_number_getter,
force=True)
Doc.set_extension('lines',
getter=TagLinkSpider.get_lines,
setter=TagLinkSpider.set_lines)
Doc.set_extension('_lines', default=list())
self.soc_spacy = socket.socket(socket.AF_INET, socket.SOCK_STREAM)
self.soc_spacy.setsockopt(socket.SOL_SOCKET, socket.SO_REUSEADDR, 1)
connect(self.soc_spacy, '', settings.SPACY_PORT)
super().__init__(**kwargs)
def time_detection(doc):
new_ents = []
for ent in doc.ents:
if ent.label_ == "CARDINAL":
next_token = doc[ent.start + 1]
if next_token.text.lower() in ("pm", "p.m", "p.m.", "am", "a.m",
"a.m."):
new_ent = Span(doc, ent.start, ent.end + 1, label="TIME")
new_ents.append(new_ent)
else:
new_ents.append(ent)
else:
new_ents.append(ent)
doc.ents = new_ents
return doc
def set_attributes(self,
schema_file: Union[str, Path] = '',
encoding: str = None) -> Set:
"""
The current version SpaCy doesn't differentiate attributes for different annotation types.
Thus, any attributes extended here will be applied to all Spans.
@param schema_file: initiate Span attributes using eHOST schema configuration file
@param encoding: text encoding
@return: a set of attribute names
"""
schema_file = self.check_file_validity(schema_file, False)
attr_names = set()
if schema_file is not None:
root = etree.parse(str(schema_file.absolute()))
for attr_def in root.iter("attributeDef"):
name = attr_def[0].text.replace(' ', '_')
default_value = attr_def[2].text
if name not in attr_names and not Span.has_extension(name):
Span.set_extension(name, default=default_value)
attr_names.add(name)
self.schema_set = True
return attr_names
def enable_spacy_extensions():
"""Enables custom extensions for spaCy for dealing with citations."""
Token.set_extension('is_in_text_citation', default=False, force=True)
Span.set_extension('tokens_without_citations',
getter=get_span_tokens_without_citations,
force=True)
Span.set_extension('text_without_citations',
getter=get_span_text_without_citations,
force=True)
Span.set_extension('text_with_ws_without_citations',
getter=get_span_text_with_ws_wo_cites,
force=True)
def test_debug_data_compile_gold_for_spans():
nlp = English()
spans_key = "sc"
pred = Doc(nlp.vocab, words=["Welcome", "to", "the", "Bank", "of", "China", "."])
pred.spans[spans_key] = [Span(pred, 3, 6, "ORG"), Span(pred, 5, 6, "GPE")]
ref = Doc(nlp.vocab, words=["Welcome", "to", "the", "Bank", "of", "China", "."])
ref.spans[spans_key] = [Span(ref, 3, 6, "ORG"), Span(ref, 5, 6, "GPE")]
eg = Example(pred, ref)
data = _compile_gold([eg], ["spancat"], nlp, True)
assert data["spancat"][spans_key] == Counter({"ORG": 1, "GPE": 1})
assert data["spans_length"][spans_key] == {"ORG": [3], "GPE": [1]}
assert data["spans_per_type"][spans_key] == {
"ORG": [Span(ref, 3, 6, "ORG")],
"GPE": [Span(ref, 5, 6, "GPE")],
}
assert data["sb_per_type"][spans_key] == {
"ORG": {"start": [ref[2:3]], "end": [ref[6:7]]},
"GPE": {"start": [ref[4:5]], "end": [ref[6:7]]},
}
def doc_findall(doc: Doc, pattern, flags=0) -> List[Span]:
return list(doc._.finditer(pattern, flags=flags))
Doc.set_extension('findall', method=doc_findall)
# # # # # # # # # # # # # # # # # # # # # # # # # # # # # #
# Span extensions for standard regex functions 'finditer' and 'findall'
# # # # # # # # # # # # # # # # # # # # # # # # # # # # # #
def span_finditer(span: Span, pattern, flags=0):
span_start_idx = span[0].idx
for m in re.finditer(pattern=pattern, string=span.text, flags=flags):
start, end = m.span()
start += span_start_idx
end += span_start_idx
yield span.doc._.idxs2span(start, end)
Span.set_extension('finditer', method=span_finditer)
def span_findall(span, pattern, flags=0):
return list(span._.finditer(pattern, flags=flags))
Span.set_extension('findall', method=span_findall)
def process_without_overlaps(self, doc: Doc,
sorted_spans: _OrderedDictItemsView,
classes: OrderedDict, attributes: OrderedDict,
relations: OrderedDict) -> Doc:
""":arg a SpaCy Doc, can be overwriten by the subclass as needed.
This function will add spans to doc.ents (defined by SpaCy as default)
which doesn't allow overlapped annotations.
@param doc: initiated SpaCy Doc
@param sorted_spans: a sorted OrderedDict Items ( spans[entity_id] = (start, end, span_text))
@param classes: a OrderedDict to map a entity id to [entity label, [attr_ids]]
@param attributes: a OrderedDict to map a attribute id to (attribute_name, attribute_value)
@param relations: a OrderedDict to map a relation_id to (label, (relation_component_ids))
@return: annotated Doc
"""
existing_entities = list(doc.ents)
new_entities = list()
# token_left_bound = 0
token_right_bound = len(doc) - 1
token_start = -1
token_end = -1
for id, span_tuple in sorted_spans:
# because SpaCy uses token offset instead of char offset to define Spans, we need to match them,
# binary search is used here to speed up
if self.store_anno_string:
start, end, span_txt = span_tuple
else:
start, end = span_tuple
# because SpaCy uses token offset instead of char offset to define Spans, we need to match them,
# binary search is used here to speed up
if start < doc[0].idx:
# If the annotation fall into a span that is before the 1st Spacy token, adjust the span to the 1st
# token
token_start = 0
token_end = 1
elif token_start >= token_right_bound:
# If the annotation fall into a span that is after the last Spacy token, adjust the span to the last
# token
token_start = token_right_bound - 1
token_end = token_right_bound
else:
token_start = self.find_start_token(start, token_start,
token_right_bound, doc)
if end >= doc[-1].idx + doc[-1].__len__():
token_end = token_right_bound + 1
else:
token_end = self.find_end_token(end, token_start,
token_right_bound, doc)
if token_start < 0 or token_start >= token_right_bound or token_end < 0 or token_end > token_right_bound:
raise ValueError(
"It is likely your annotations overlapped, which process_without_overlaps doesn't support parsing "
"those. You will need to initiate the EhostDocReader with 'support_overlap=True' in the arguements"
)
if token_start >= 0 and token_end > 0:
span = Span(doc, token_start, token_end, label=classes[id][0])
for attr_id in classes[id][1]:
if attr_id not in attributes:
continue
attr_name = attributes[attr_id][0]
attr_value = attributes[attr_id][1]
setattr(span._, attr_name, attr_value)
if self.store_anno_string and span_txt is not None:
setattr(span._, "span_txt", span_txt)
new_entities.append(span)
token_start = token_end
else:
raise OverflowError(
'The span of the annotation: {}[{}:{}] is out of document boundary.'
.format(classes[id][0], start, end))
pass
doc.ents = existing_entities + new_entities
return doc
span_attn_getter,
span_nctokens_getter,
span_wp2ncid_getter,
span_wp2tokid_getter,
span_wp_getter,
span_wp_slice_getter,
)
Doc.set_extension("wp2ncid", getter=doc_wp2ncid_getter)
Doc.set_extension("nctokens", getter=doc_nctokens_getter)
Doc.set_extension("tokid2nc", getter=doc_tokid2nc_getter)
Doc.set_extension("wp2tokid", getter=doc_wp2tokid_getter)
Doc.set_extension("tokid2ncid", getter=doc_tokid2ncid_getter)
Doc.set_extension("tokid2wp", getter=doc_tokid2wp_getter)
Span.set_extension("wp_slice", getter=span_wp_slice_getter)
Span.set_extension("wp2tokid", getter=span_wp2tokid_getter)
Span.set_extension("attention", getter=span_attn_getter)
Span.set_extension("wordpieces", getter=span_wp_getter)
Span.set_extension("wp2ncid", getter=span_wp2ncid_getter)
Span.set_extension("nctokens", getter=span_nctokens_getter)
def load_danish(spacy_model: str = "da_core_news_sm",
transformer: str = "Maltehb/danish-bert-botxo"):
nlp = spacy.load(spacy_model)
if transformer:
# add transformer
# Construction via add_pipe with custom config
config = {
def f(doc: Doc) -> Doc:
doc.ents += (Span(doc, 3, 5, "foo"), )
return doc
def _pseudofy_side(self,
rel: brat_data.Relation,
sentence: Span,
k: int,
do_left=True) -> SentenceGenerator:
rel = _make_contig_rel(rel)
if not rel:
return
# _make_contig_rel does make a deep copy but no interesting changes have been made yet
logging.info(f'Original instance: {rel}')
ent, other_ent = (rel.arg1, rel.arg2) if do_left else (rel.arg2,
rel.arg1)
text = str(sentence)
span_start = sentence.start_char
start, end = adjust_spans(ent, -span_start)
adjust_spans(other_ent, -span_start)
try:
original_pos = [t.pos_ for t in sentence.char_span(start, end)]
except TypeError:
# The char span doesn't line up with any tokens,
# thus we can't figure out if the prediction is the right POS
logging.info(
'Instance rejected; the spans given do not align with tokens according to the spaCy model'
)
return None
masked_sentence = text[:start] + '[MASK]' + text[end:]
tokenized_sentence = self.bert_tokenizer.tokenize(masked_sentence)
indexed_tokens = self.bert_tokenizer.convert_tokens_to_ids(
tokenized_sentence)
token_tensor = torch.tensor(indexed_tokens)
mask_tensor = torch.tensor(
[token != '[MASK]' for token in tokenized_sentence],
dtype=torch.float)
if len(token_tensor) > 512:
# This is the token limit we report on, but the limit depends on the BERT model
return None
with torch.no_grad():
result = self.bert(token_tensor.unsqueeze(0),
mask_tensor.unsqueeze(0),
masked_lm_labels=None)
result = result[0].squeeze(0)
scores = torch.softmax(result, dim=-1)
mask_index = tokenized_sentence.index('[MASK]')
topk_scores, topk_indices = torch.topk(scores[mask_index, :],
k,
sorted=True)
topk_tokens = self.bert_tokenizer.convert_ids_to_tokens(topk_indices)
for token, score in zip(topk_tokens, topk_scores):
new_sent = text[:start] + token + text[end:]
new_doc = self.nlp(new_sent)
new_span = new_doc.char_span(start, start + len(token))
if new_span is None:
continue
pos_match = [t.pos_ for t in new_span] == original_pos
this_rel = deepcopy(rel)
ent, other_ent = (this_rel.arg1,
this_rel.arg2) if do_left else (this_rel.arg2,
this_rel.arg1)
ent.spans = [(start, start + len(token))]
if ent.start < other_ent.start:
# If the entity being changed comes before the one not being changed, the spans of the other must
# also be adjusted; it is not guaranteed that `rel.arg1` always comes before `rel.arg2`
new_offset = len(token) - len(ent.mention)
adjust_spans(other_ent, new_offset)
ent.mention = token
new_ps = PseudoSentence(this_rel, new_sent, float(score),
pos_match)
logging.info(f'New instance: {new_ps}')
yield new_ps
def process_support_overlaps(self, doc: Doc,
sorted_spans: _OrderedDictItemsView,
classes: OrderedDict, attributes: OrderedDict,
relations: OrderedDict) -> Doc:
""":arg a SpaCy Doc, can be overwriten by the subclass as needed.
This function will add spans to doc._.concepts (defined in 'read' function above,
which allows overlapped annotations.
@param doc: initiated SpaCy Doc
@param sorted_spans: a sorted OrderedDict Items ( spans[entity_id] = (start, end, span_text))
@param classes: a OrderedDict to map a entity id to [entity label, [attr_ids]]
@param attributes: a OrderedDict to map a attribute id to (attribute_name, attribute_value)
@param relations: a OrderedDict to map a relation_id to (label, (relation_component_ids))
@return: annotated Doc
"""
existing_concepts: dict = doc._.concepts
# token_left_bound = 0
previous_abs_end = 0
token_right_bound = len(doc) - 1
token_start = -1
token_end = -1
for id, span_tuple in sorted_spans:
# because SpaCy uses token offset instead of char offset to define Spans, we need to match them,
# binary search is used here to speed up
if self.store_anno_string:
start, end, span_txt = span_tuple
else:
start, end = span_tuple
if start < doc[0].idx:
# If the annotation fall into a span that is before the 1st Spacy token, adjust the span to the 1st
# token
token_start = 0
token_end = 1
elif token_start >= token_right_bound:
# If the annotation fall into a span that is after the last Spacy token, adjust the span to the last
# token
self.logger.debug(
"token_start {} >= token_right_bound {}".format(
token_start, token_right_bound))
token_start = token_right_bound
token_end = token_right_bound + 1
else:
# if start < previous_abs_end:
# self.logger.debug("To find {} between token_start - 1({}[{}]) and token_right_bound({}[{}])"
# .format(start, token_start-1, doc[token_start-1].idx,
# token_right_bound, doc[token_right_bound].idx), )
# token_start = self.find_start_token(start, token_start - 1 if token_start > 0 else 0,
# token_right_bound, doc)
# self.logger.debug('\tfind token_start={}[{}]'.format(token_start, doc[token_start].idx))
#
# else:
self.logger.debug(
"To find {} between token_start ({}[{}]) and token_right_bound({}[{}])"
.format(start, token_start, doc[token_start].idx,
token_right_bound, doc[token_right_bound].idx), )
token_start = self.find_start_token(start, token_start,
token_right_bound, doc)
self.logger.debug("\tfind start token {}('{}')".format(
token_start, doc[token_start]))
if end >= doc[-1].idx + doc[-1].__len__():
self.logger.debug(
"end ({}) >= doc[-1].idx ({}) + doc[-1].__len__() ({})"
.format(end, doc[-1].idx, doc[-1].__len__()))
token_end = token_right_bound + 1
else:
self.logger.debug(
"To find token_end starts from {} between token_start ({}[{}]) and token_right_bound({}[{}])"
.format(end, token_start, doc[token_start].idx,
token_right_bound, doc[token_right_bound].idx))
token_end = self.find_end_token(end, token_start,
token_right_bound, doc)
self.logger.debug("\tFind end token {}('{}')".format(
token_end, doc[token_end]))
if token_start >= 0 and token_end > 0:
span = Span(doc, token_start, token_end, label=classes[id][0])
if self.logger.isEnabledFor(logging.DEBUG):
import re
if re.sub('\s+', ' ', span._.span_txt) != re.sub(
'\s+', ' ', str(span)):
self.logger.debug('{}[{}:{}]\n\t{}<>\n\t{}<>'.format(
classes[id][0], token_start, token_end,
re.sub('\s+', ' ', span._.span_txt),
re.sub('\s+', ' ', str(span))))
for attr_id in classes[id][1]:
if attr_id not in attributes:
continue
attr_name = attributes[attr_id][0]
attr_value = attributes[attr_id][1]
setattr(span._, attr_name, attr_value)
if self.store_anno_string and span_txt is not None:
setattr(span._, "span_txt", span_txt)
if classes[id][0] not in existing_concepts:
existing_concepts[classes[id][0]] = list()
existing_concepts[classes[id][0]].append(span)
# token_start = token_end
previous_abs_end = token_start
else:
raise OverflowError(
'The span of the annotation: {}[{}:{}] is out of document boundary.'
.format(classes[id][0], start, end))
pass
return doc
#print("*) PRINT Spacy NER Formatted string - 02")
#for i in range(0,len(ner_ita_spacy_doc.ents)):
# ner_ita_phrase_01_ner_formatted = ner_ita_phrase_01_ner_formatted.replace(ner_ita_spacy_doc.ents[i].text, ner_ita_spacy_doc.ents[i].text+"("+ner_ita_spacy_doc.ents[i].label_+")")
print("*) Spacy ENTITY CORRECTION - ITA")
'''
Obtaing hash of labels with spacy
'''
money_tag = ner_ita_spacy_doc.vocab.strings["MONEY"]
print(money_tag)
print("*) Original phrase")
print(ner_ita_phrase_01)
print(ner_ita_spacy_doc.ents)
## IMPO: SPACY DEFINE AN ADHOC NER ENTITY - MONEY
it_money_ner = Span(ner_ita_spacy_doc, 14, 18, label=money_tag)
print(it_money_ner)
ner_ita_spacy_doc.ents = list(ner_ita_spacy_doc.ents) + [it_money_ner]
print(ner_ita_spacy_doc.ents)
ner_ita_phrase_01_ner_formatted = ner_ita_phrase_01
#for i in range(0,len(ner_ita_spacy_doc.ents)):
# ner_ita_phrase_01_ner_formatted = ner_ita_phrase_01_ner_formatted.replace(ner_ita_spacy_doc.ents[i].text, ner_ita_spacy_doc.ents[i].text+"("+ner_ita_spacy_doc.ents[i].label_+")")
ner_ita_phrase_01_ner_formatted = ner_ita_phrase_01
for i in range(0, len(ner_ita_spacy_doc.ents)):
ner_ita_phrase_01_ner_formatted = ner_ita_phrase_01_ner_formatted.replace(
ner_ita_spacy_doc.ents[i].text, ner_ita_spacy_doc.ents[i].text + "(" +
ner_ita_spacy_doc.ents[i].label_ + ")")
print("*) PRINT Spacy NER Formatted string - 01 - + new ENTITY - MONEY")
print(ner_ita_phrase_01_ner_formatted)
import ahocorasick
import spacy
import textspan
from spacy.tokens import Doc
from spacy.tokens.span import Span
from spacy.util import filter_spans
from typing_extensions import Literal
from camphr.utils import SerializationMixin, get_doc_char_span
# Sometimes matched text is different from original text
# since `PatternSearcher` can match the `lemma`.
# This extension holds the matched text.
PATTERN_MATCH_AS = "pattern_match_as"
Span.set_extension(PATTERN_MATCH_AS, default=None, force=True)
@spacy.component("pattern_searcher")
class PatternSearcher(SerializationMixin):
serialization_fields = [
"model",
"label_type",
"custom_label",
"custom_label_map",
"destructive",
"lemma",
"lower",
"cfg",
"normalizer",
]
def __call__(self, doc):
#ToDo Could use refactoring for readability and structure, as it's too complex now
full_name_spans = []
min_span_size = 2
max_span_size = 2
is_first_char_capped_list = [
doc[i].text[0].isupper() for i in range(len(doc))
]
i = 0
while i < len(doc) - 1:
is_first_name_this = doc[i]._.get(self.first_name_extension_name)
is_last_name_next = doc[i + 1]._.get(self.last_name_extension_name)
both_capped = is_first_char_capped_list[
i] and is_first_char_capped_list[i + 1]
if is_first_name_this and is_last_name_next and both_capped:
doc[i]._.set(self.token_extension_name, self.ANOT_INIT)
doc[i + 1]._.set(self.token_extension_name, self.ANOT_OTHER)
new_span = Span(doc, i, i + 1, label=self.SPAN_LABEL)
span_start = i
span_end = i + 1
#look-back for more first names
first_first_name = i
while first_first_name-1>= 0 and doc[first_first_name-1]._.get(self.first_name_extension_name)\
and is_first_char_capped_list[first_first_name-1]:
first_first_name -= 1
span_start = first_first_name
if first_first_name != i: #name starts earlier
doc[i]._.set(self.token_extension_name, self.ANOT_OTHER)
doc[first_first_name]._.set(self.token_extension_name,
self.ANOT_INIT)
for j in range(first_first_name + 1, i):
doc[first_first_name]._.set(self.token_extension_name,
self.ANOT_OTHER)
# Check if considering the last name as a first name would still yield a valid match
# when adding the following word
look_ahead_counter = 1
while (i+1+look_ahead_counter) < len(doc) \
and doc[i+look_ahead_counter]._.get(self.first_name_extension_name) \
and doc[i+1+look_ahead_counter]._.get(self.last_name_extension_name)\
and is_first_char_capped_list[i+1+look_ahead_counter]:
doc[i + 1 + look_ahead_counter]._.set(
self.token_extension_name, self.ANOT_OTHER)
look_ahead_counter += 1
span_end += 1
#Check if it can be extended with more last names
while (span_end + 1 < len(doc) and doc[span_end + 1]._.get(
self.last_name_extension_name)
and is_first_char_capped_list[span_end + 1]):
doc[span_end + 1]._.set(self.token_extension_name,
self.ANOT_OTHER)
span_end += 1
full_name_spans.append(new_span)
i = span_end + 1
else:
i += 1
doc._.set(self.doc_extension_name, full_name_spans)
return doc
def __init__(self, txt_path):
super(ASK, self).__init__()
f_path = txt_path
with open(f_path, 'r') as f:
txt = f.readlines()
txt = [x.strip() for x in txt]
sentence_regex = "(?<!\w\.\w.)(?<![A-Z][a-z]\.)(?<=\.|\?)\s"
regex = re.compile(sentence_regex)
txt = [removeParentheses(x) for x in txt if regex.search(x)]
txt = ''.join(str(elem) for elem in txt)
self.txt = txt
nlp = spacy.load('en')
doc = nlp(unicode(txt))
questions = []
words = []
for word in doc:
if word.pos_ == "NOUN" or word.pos_ == "VERB":
words.append(word.lemma_)
count = Counter(words)
for sent in doc.sents:
rel = 0
for word in sent:
rel += count.get(word.lemma_, 0)
rel = rel/float(len(sent))
if sent[0].text == "This":
remain = Span(doc, sent.start+1, sent.end-1)
questions.append(("What " + remain.text + "?", rel+10))
# What questions
if (sent.root.lemma_ == "be"):
for r in sent.root.rights:
join = ""
first = True
for w in r.subtree:
if first and w.ent_type_ == "":
join += ' ' + w.lower_
else:
first = False
join += ' ' + w.text
if join[-1] == ",":
join = join[:-2]
what_question_1 = "What " + sent.root.text + " " + join + "?"
questions.append((what_question_1, rel))
break
for r in sent.root.lefts:
join = ""
first = True
for w in r.subtree:
if first and w.ent_type_ == "":
join += ' ' + w.lower_
else:
first = False
join += ' ' + w.text
if join[-1] == ",":
join = join[:-2]
what_question_2 = "What " + sent.root.text + " " + join + "?"
questions.append((what_question_2, rel))
# Who questions
subject = ["he", "she"]
for i in range(0, len(sent)-1) :
if sent[i].dep_ == "nsubj" and sent[i].ent_type_ == "PERSON" or sent[i].text in subject:
if i > 0:
start = Span(doc, sent.start, i+sent.start-1)
else:
start = Span(doc, sent.start, sent.start)
i = i+1
while i < len(sent)-1:
if sent[i].ent_type_ == "PERSON":
i = i+1
elif sent[i].dep_ == "nsubj":
i = i+1
else:
break
end = Span(doc, i + sent.start, sent.end-1)
if (len(start) == 0):
who_question = "Who " + end.text + "?"
questions.append((who_question, rel))
else:
who_question = "Who " + end.text + "?"
questions.append((who_question, rel))
break
# When questions
# Only works when original sentence is "In ____, blah blah."
for i in range(0, len(sent)-1):
if (sent[i].ent_type_ == "DATE" and sent[i].pos_ != "ADJ"):
hi = Span(doc, sent.start, i+sent.start)
head = sent[i].head
while i < len(sent) - 1 and (sent[i].ent_type_ == "DATE" or sent[i].pos_ == "PUNCT"):
i = i+1
end = Span(doc, i + sent.start, sent.end-1)
verb = sent[i]
for t in sent.root.lefts:
verb = t
if verb.lemma_ == "be":
final = "When was "
else:
final = "When did "
if len(end) > 0 and (end[0].pos_ == "NOUN" or end[0].pos_ == "DET"):
for token in end:
if verb.lemma_ == "be" and token.lemma_ == "be":
final = final
elif verb.lemma_ != "be" and token == sent.root:
final = final + sent.root.lemma_ + " "
else:
final = final + token.orth_ + " "
when_question_1 = final[:-1] + "?"
questions.append((when_question_1, rel))
break
for i in range(0, len(sent)-1):
if (sent[i].ent_type_ == "DATE"):
if (i > 0):
front = Span(doc, sent.start, i+sent.start-1)
else:
font = []
valid = False
while i < len(sent) - 1 and (sent[i].ent_type_ == "DATE" or sent[i].pos_ == "PUNCT"):
i = i+1
valid = True
if valid:
end = Span(doc, sent.start + i, sent.end-1)
verb = sent[i]
for t in sent.root.lefts:
verb = t
if verb.lemma_ == "be":
final = "When was "
else:
final = "When did "
if len(end) > 0 and (end[0].pos_ == "NOUN" or end[0].pos_ == "DET"):
for token in front:
if verb.lemma_ == "be" and token.lemma_ == "be":
final = final
elif verb.lemma_ != "be" and token == sent.root:
final = final + sent.root.lemma_ + " "
elif token == front[0] and token.ent_type_ == "":
final = final + token.lower_ + " "
else:
final = final + token.orth_ + " "
for token in end:
if verb.lemma_ == "be" and token.lemma_ == "be":
final = final
elif verb.lemma_ != "be" and token == sent.root:
final = final + sent.root.lemma_ + " "
else:
final = final + token.orth_ + " "
when_question_2 = final[:-1] + "?"
questions.append((when_question_2, rel))
break
# Where questions
wheretag = ["GPE", "LOC", "FACILTY", "ORG"]
for i in range(0, len(sent)-1):
if (sent[i].ent_type_ in wheretag) and (sent[i-1].ent_type_ not in wheretag) and (sent[i-1].tag_ == "IN"):
oneloc = " " + str(sent[i-1])+ " " + str(sent[i])
j = i
while j < len(sent)-1 and sent[j+1].ent_type_ in wheretag:
oneloc += " " + str(sent[j+1])
j = j + 1
i = j
where_question1 = ""
where_question2 = ""
for k in range(0, len(sent)-1):
if sent[k] == sent.root and (sent[k-1].pos_ == 'VERB'):
where_question1 = "Where" + " " + two_verbs(k, sent, doc)[0]
where_question2 = "Where" + " " + two_verbs(k, sent, doc)[1]
elif sent[k] == sent.root:
where_question1 = "Where" + " " + one_verb(k, sent, doc)[0]
where_question2 = "Where" + " " + one_verb(k, sent, doc)[1]
where_question1 = where_question1.replace(oneloc, "")
where_question2 = where_question2.replace(oneloc, "")
questions.append((where_question1, rel))
questions.append((where_question2, rel))
break
# DO/ DID DOES HAVE WAS IS questions
for i in range(0, len(sent)-1):
vquest1 = ""
vquest2 = ""
if sent[i] == sent.root and (sent[i-1].pos_ == 'VERB'):
vquest1 = (two_verbs(i, sent, doc)[0])
vquest1 = vquest1[0].capitalize() + vquest1[1:]
vquest2 = (two_verbs(i, sent, doc)[1])
vquest2 = vquest2[0].capitalize() + vquest2[1:]
questions.append((vquest1, rel))
questions.append((vquest2, rel))
break
elif sent[i] == sent.root:
vquest1 = (one_verb(i, sent, doc)[0])
vquest1 = vquest1[0].capitalize() + vquest1[1:]
vquest2 = (one_verb(i, sent, doc)[1])
vquest2 = vquest2[0].capitalize() + vquest2[1:]
questions.append((vquest1, rel))
questions.append((vquest2, rel))
break
pronouns = ["he", "she", "his", "her"]
questions = sorted(questions, key=lambda x: x[1], reverse=True)
goodQuestions = []
goodPronouns = []
for q in questions:
if (q[0].count(' ') > 3):
sentence = q[0]
try:
matches = tool.check(sentence)
if len(matches)==0:
words = q[0].split()
if not any(p == w for w in words for p in pronouns):
goodQuestions.append(sentence)
else:
goodPronouns.append(sentence)
except:
words = q[0].split()
if not any(p == w for w in words for p in pronouns):
goodQuestions.append(sentence)
else:
goodPronouns.append(sentence)
count = 0
for i in range(0, 4):
for j in range(0, len(goodQuestions)/4):
if count < qNum:
count = count+1
print goodQuestions[j*4+i]
for i in range(0, 4):
for j in range(0, len(goodPronouns)/4):
if count < qNum:
count = count+1
print goodPronouns[j*4+i]
def __init__(self, txt_path):
super(NER, self).__init__()
cur_path = os.path.dirname(__file__)
rel_path = '../data/' + txt_path
f_path = os.path.join(cur_path, rel_path)
with open(f_path, 'r') as f:
txt = f.readlines()
txt = [x.strip() for x in txt]
sentence_regex = "(?<!\w\.\w.)(?<![A-Z][a-z]\.)(?<=\.|\?)\s"
regex = re.compile(sentence_regex)
txt = [x for x in txt if regex.search(x)]
txt = ''.join(str(elem) for elem in txt)
self.txt = txt
nlp = spacy.load('en')
doc = nlp(unicode(txt))
questions = []
when = []
words = []
for word in doc:
if word.pos_ == "NOUN" or word.pos_ == "VERB":
words.append(word.lemma_)
count = Counter(words)
for sent in doc.sents:
rel = 0
for word in sent:
rel += count.get(word.lemma_, 0)
rel = rel/float(len(sent))
# What questions
if (sent.root.lemma_ == "be"):
for r in sent.root.rights:
join = ' '.join(w.text for w in r.subtree)
if join[-1] == ",":
join = join[:-2]
what_question_1 = "What " + sent.root.text + " " + join + "?"
questions.append((format_question(what_question_1), rel+gscore(what_question_1)))
break
for r in sent.root.lefts:
join = ' '.join(w.text for w in r.subtree)
if join[-1] == ",":
join = join[:-2]
what_question_2 = "What " + sent.root.text + " " + join + "?"
questions.append((format_question(what_question_2), rel+gscore(what_question_2)))
# Who questions
subject = ["he", "she"]
for i in range(0, len(sent)-1) :
if sent[i].dep_ == "nsubj" and sent[i].ent_type_ == "PERSON" or sent[i].text in subject:
if i > 0:
start = Span(doc, sent.start, i+sent.start-1)
else:
start = Span(doc, sent.start, sent.start)
i = i+1
while i < len(sent)-1:
if sent[i].ent_type_ == "PERSON":
i = i+1
elif sent[i].dep_ == "nsubj":
i = i+1
else:
break
end = Span(doc, i + sent.start, sent.end-1)
if (len(start) == 0):
# print "Who " + end.text + "?"
who_question = "Who " + end.text + "?"
questions.append((format_question(who_question), rel+gscore(who_question)))
else:
# print start.text + " who " + end.text + "?"
who_question = start.text + " who " + end.text + "?"
questions.append((format_question(who_question), rel+gscore(who_question)))
break
# When questions
# Only works when original sentence is "In ____, blah blah."
for i in range(0, len(sent)-1):
if (sent[i].ent_type_ == "DATE" and sent[i].pos_ != "ADJ"):
hi = Span(doc, sent.start, i+sent.start)
head = sent[i].head
while i < len(sent) - 1 and (sent[i].ent_type_ == "DATE" or sent[i].pos_ == "PUNCT"):
i = i+1
end = Span(doc, i + sent.start, sent.end-1)
verb = sent[i]
for t in sent.root.lefts:
verb = t
if verb.lemma_ == "be":
final = "When was "
else:
final = "When did "
for token in end:
if verb.lemma_ == "be" and token.lemma_ == "be":
final = final
elif verb.lemma_ != "be" and token == sent.root:
final = final + sent.root.lemma_ + " "
else:
final = final + token.orth_ + " "
# print final[:-1] + "?"
when_question_1 = final[:-1] + "?"
questions.append((when_question_1, rel+gscore(when_question_1)))
when.append((when_question_1, rel+gscore(when_question_1)))
break
for i in range(0, len(sent)-1):
if (sent[i].ent_type_ == "DATE"):
if (i > 0):
front = Span(doc, sent.start, i+sent.start-1)
else:
font = []
valid = False
while i < len(sent) - 1 and (sent[i].ent_type_ == "DATE" or sent[i].pos_ == "PUNCT"):
i = i+1
valid = True
if valid:
end = Span(doc, sent.start + i, sent.end-1)
verb = sent[i]
for t in sent.root.lefts:
verb = t
if verb.lemma_ == "be":
final = "When was "
else:
final = "When did "
for token in front:
if token.ent_type_ == "":
final = final + token.lower_ + " "
else:
final = final + token.text + " "
for token in end:
if verb.lemma_ == "be" and token.lemma_ == "be":
final = final
elif verb.lemma_ != "be" and token == sent.root:
final = final + sent.root.lemma_ + " "
else:
final = final + token.orth_ + " "
# print final[:-1] + "?"
when_question_2 = final[:-1] + "?"
questions.append((when_question_2, rel+gscore(when_question_2)))
when.append((when_question_2, rel+gscore(when_question_2)))
break
# Where questions
for i in range(0, len(sent)-1):
# if (sent[i].ent_type_ == "GPE" and sent[i-1].ent_type_ != "GPE"):
if (sent[i].ent_type_ == "GPE" and sent[i-1].ent_type_ != "GPE" and sent[i-1].tag_ == "IN"):
# print sent
oneloc = str(sent[i])
j = i
while j < len(sent)-1 and sent[j+1].ent_type_ == "GPE":
oneloc += " " + str(sent[j+1])
j = j + 1
# possible_locations.append(oneloc)
i = j
for r in sent.root.rights:
if sent.root.lemma_ == "be":
final = "Where was "
elif sent.root.tag_ == "VB":
final = "Where do "
elif sent.root.tag_ == "VBP":
final = "Where have "
else:
final = "Where did "
subject = ''
for l in sent.root.lefts:
if l.right_edge.dep_=='nsubj':
subject = str(l.right_edge)
break
else:
subject = ' '.join(w.text for w in l.subtree)
final += subject + " " + sent.root.lemma_ + " " + ' '.join(w.text for w in r.subtree) + "?"
break
final = final.replace(oneloc, "")
where_question = final[:-1] + "?"
questions.append((where_question, rel+gscore(where_question)))
# print ""
break
# DO/ DID DOES HAVE questions
if (sent.root.pos_ == "VERB"):
# print sent.root.tag_
v_question = ""
for r in sent.root.rights:
if sent.root.tag_ == 'VB':
v_question = 'Do'
elif sent.root.tag_ == 'VBD':
v_question = 'Did'
elif sent.root.tag_ == 'VBZ':
v_question = 'Does'
elif sent.root.tag_ == 'VBP':
v_question = 'Have'
elif sent.root.tag_ == 'VBN':
break
else:
break
subject =''
for l in sent.root.lefts:
if l.right_edge.dep_=='nsubj':
subject = str(l.right_edge)
break
elif v_question == 'Have':
subject = ' '.join(w.text for w in l.subtree)
break
else: subject = ' '.join(w.text for w in l.subtree)
v_question += " " + subject + " " + sent.root.lemma_ + " " +' '.join(w.text for w in r.subtree) + '?'
break
# print v_question
questions.append((v_question, rel+gscore(v_question)))
# print ""
# Is Was Were Questions
if (sent.root.lemma_ == "be"):
for r in sent.root.rights:
if sent.root.text == '\'s':
isquest1 == 'Is'
else:
isquest1 = sent.root.text.capitalize()
for l in sent.root.lefts:
isquest1 += " " + ' '.join(w.text for w in l.subtree)
isquest1 += ' ' +' '.join(w.text for w in r.subtree) +'?'
break
# print isquest1
questions.append((isquest1, rel+gscore(isquest1)))
# print ""
questions = sorted(questions, key=lambda x: x[1], reverse=True)
goodQuestions = []
for q in questions:
if q[0].count(' ') > 3:
sentence = q[0]
matches = tool.check(sentence)
if len(matches) == 0:
goodQuestions.append(q[0])
for i in range(0, qNum):
if (i*4) < len(goodQuestions)
print goodQuestions[i*4]