def create_parsed_sentences(doc, separate_sentences=True):
if len(doc) == 0:
return [ParsedSentence('', [])]
sentences = []
begin = 0
base_idx = doc[0].idx
if separate_sentences:
for token in doc:
if token.sent_start and token.i > 0:
end = token.i
morphs = [
Morph(
i,
t.idx - base_idx,
t.orth_,
t.lemma_,
t.pos_,
t.tag_,
ex_attr(t).inf,
t.whitespace_,
) for i, t in enumerate(doc[begin:end])
]
for m, t in zip(morphs, doc[begin:end]):
m.dep_morph = morphs[t.head.i - begin]
m.dep_label = t.dep_.lower()
sentences.append(
ParsedSentence(doc.text[base_idx:token.idx], morphs))
begin = end
base_idx = token.idx
if begin < len(doc):
morphs = [
Morph(
i,
t.idx - base_idx,
t.orth_,
t.lemma_,
t.pos_,
t.tag_,
ex_attr(t).inf,
t.whitespace_,
) for i, t in enumerate(doc[begin:])
]
for m, t in zip(morphs, doc[begin:]):
m.dep_morph = morphs[t.head.i - begin]
m.dep_label = t.dep_.lower()
sentences.append(ParsedSentence(doc.text[base_idx:], morphs))
return sentences
def rewrite_with_tokens(self, rewriting_morph_index, tokens):
origin = self.morphs[rewriting_morph_index]
origin_pos = origin.pos
t = tokens[0]
origin.surface = t.orth_
origin.lemma = t.lemma_
origin.pos = t.pos_
origin.tag = t.tag_
origin.inf = ex_attr(t).inf
origin.trailing_space = t.whitespace_
if origin_pos != origin.pos:
origin.dep_label = '{}_as_{}'.format(origin.dep_label, origin_pos)
if len(tokens) == 1:
return
label = 'as_{}'.format(origin.pos)
others = [
Morph(
rewriting_morph_index + i + 1,
origin.offset + t.idx - tokens[0].idx,
t.orth_,
t.lemma_,
t.pos_,
t.tag_,
ex_attr(t).inf,
t.whitespace_,
) for i, t in enumerate(tokens[1:])
]
offset = origin.offset
if origin.trailing_space:
offset += 1
for m in others:
m.offset = offset
offset += len(m.surface)
if m.trailing_space:
offset += 1
m.dep_morph = origin
m.dep_label = label
for m in self.morphs[rewriting_morph_index + 1:]:
m.id += len(others)
self.morphs[rewriting_morph_index + 1:rewriting_morph_index +
1] = others
def to_doc(self, vocab, is_parsed=False):
words = [morph.surface for morph in self.morphs]
spaces = [morph.trailing_space for morph in self.morphs]
doc = Doc(vocab, words=words, spaces=spaces)
for token, morph in zip(doc, self.morphs):
token.tag_ = morph.tag
token.pos_ = morph.pos
ex_attr(token).inf = morph.inf
token.lemma_ = morph.lemma # work around: lemma_ must be set after tag_ (spaCy's bug)
if is_parsed and morph.dep_label:
token.dep_ = morph.dep_label
token.head = doc[morph.dep_morph.id]
return doc
def unify_range(gold_tokens, start, end, replacing_token, extend_dep_labels):
dep_outer_id = None
dep_outer_label = None
head_pos = None
for g in gold_tokens[start:end]:
head_id = g['id'] + g['head']
if head_id < start or end <= head_id or g['head'] == 0:
if dep_outer_id is None:
dep_outer_id = head_id
dep_outer_label = g['dep']
head_pos = g['pos']
elif dep_outer_id != head_id:
return False
if dep_outer_id is None:
print(gold_tokens[start:end], file=sys.stderr)
raise Exception('unexpected state')
elif start < dep_outer_id < end:
dep_outer_id = start
g = gold_tokens[start]
g['orth'] = replacing_token.orth_
g['lemma'] = replacing_token.lemma_
g['pos'] = replacing_token.pos_
g['tag'] = replacing_token.tag_
g['inf'] = ex_attr(replacing_token).inf
g['whitespace'] = replacing_token.whitespace_ != ''
g['head'] = dep_outer_id - start
if dep_outer_label.startswith('as_'):
g['dep'] = dep_outer_label
else:
dep = dep_outer_label.split('_as_')[0]
g['dep'] = dep if not extend_dep_labels or head_pos == g[
'pos'] else '{}_as_{}'.format(dep, head_pos)
for g in gold_tokens:
if g['id'] <= start and end <= g['id'] + g['head']:
g['head'] -= end - start - 1
elif g['id'] <= start < g['id'] + g['head']:
g['head'] = start - g['id']
elif g['id'] + g['head'] <= start and end <= g['id']:
g['head'] += end - start - 1
elif g['id'] + g['head'] < end <= g['id']:
g['head'] = end - g['id'] - 1
for g in gold_tokens[end:]:
g['id'] -= end - start - 1
del gold_tokens[start + 1:end]
return True
def unify_range(self, start, end, replacing_token):
dep_outer_id = None
dep_outer_label = None
head = None
for m in self.morphs[start:end]:
if start <= m.dep_morph.id < end:
if m.dep_morph.id == m.id:
if dep_outer_id:
return False
else:
dep_outer_id = m.id
dep_outer_label = m.dep_label
head = m
elif dep_outer_id:
if dep_outer_id == m.dep_morph.id:
head = m
else:
return False
else:
dep_outer_id = m.dep_morph.id
dep_outer_label = m.dep_label
head = m
if dep_outer_id is None:
raise Exception('unexpected state')
elif start < dep_outer_id < end:
dep_outer_id = start
origin = self.morphs[start]
origin.surface = replacing_token.orth_
origin.lemma = replacing_token.lemma_
origin.pos = replacing_token.pos_
origin.tag = replacing_token.tag_
origin.inf = ex_attr(replacing_token).inf
origin.trailing_space = replacing_token.whitespace_
origin.dep_morph = self.morphs[dep_outer_id]
origin.dep_label = dep_outer_label if origin.pos == head.pos else '{}_as_{}'.format(
dep_outer_label, head.pos)
for m in self.morphs:
if start < m.dep_morph.id < end:
m.dep_morph = origin
del self.morphs[start + 1:end]
for m in self.morphs:
if m.id >= end:
m.id -= end - start - 1
return True
