def _process_internal(question):
assert(isinstance(question, spacy.tokens.span.Span))
if len(question) < 2:
return str(question) + "@placeholder"
question = "Which " + str(question) + " ?"
nlp = SLoader.get_full_spacy_nlp()
question = list(nlp(question).sents)[0]
qtype = get_question_type(question)
if qtype == QType.WHICH_OF:
return which_of.process(question)
if qtype == QType.IN_WHICH_OF:
# No examples in the entire SQuAD dataset.
return in_which_of.process(question)
if qtype == QType.WHICH_NOUN:
return which_noun.process(question)
if qtype == QType.WHICH_BE:
return which_be.process(question)
if qtype == QType.IN_WHICH_NOUN:
# No examples in the entire SQuAD dataset.
return in_which_noun.process(question)
if qtype == QType.WHICH_VERB:
return which_verb.process(question)
question = question[1:] # Remove "Which"
question = ["@placeholder"] + [str(x) for x in question]
question[-1] = '.' # Replace "?" with "."
return ' '.join(question)
def _process_internal(question):
assert (isinstance(question, spacy.tokens.span.Span))
if len(question) < 2:
return str(question) + "@placeholder"
question = "What " + str(question) + " ?"
nlp = SLoader.get_full_spacy_nlp()
question = list(nlp(question).sents)[0]
qtype = get_question_type(question)
if qtype == QType.WHAT_BE:
return what_be.process(question)
if qtype == QType.WHAT_DO:
return what_do.process(question)
if qtype == QType.IN_WHAT:
return in_what.process(question)
if qtype == QType.WHAT_NOUN:
return what_noun.process(question)
if qtype == QType.WHAT_VERB:
return what_verb.process(question)
question = question[1:] # Remove "What"
question = ["@placeholder"] + [str(x) for x in question]
question[-1] = '.' # Replace "?" with "."
return ' '.join(question)
def process(question):
assert (isinstance(question, spacy.tokens.span.Span))
assert (len(question) >= 3)
skipped, question = _split_question(question)
assert (isinstance(skipped, list))
assert (isinstance(question, spacy.tokens.span.Span))
if len(skipped) >= 1:
skipped[0] = "in" # Not "In".
while len(question) >= 1 and question[-1].is_punct:
question = question[:-1]
question = [str(x) for x in question]
if not question[0].isupper():
question[0] = question[0].capitalize()
skipped = ' '.join(skipped)
question = ' '.join(question)
# Swap main question and "In ...".
question = question + ", " + skipped + " ?"
nlp = SLoader.get_full_spacy_nlp()
doc = nlp(question)
question = doc[0:len(doc)] # Convert to spaCy Span (not Doc).
return _process_internal(question)
def split_in_sentences(text):
assert (isinstance(text, str))
nlp = SLoader.get_full_spacy_nlp()
return list(nlp(text).sents)