class NegexTriggerTagger:
def __init__(self, rules: List[List[str]] = None, tokens_range: int = 40):
if rules is None:
with (Path(__file__).parent / 'negex_triggers.txt').open('r') as f:
rules = make_rules(f)
self.dawg = DAWG()
for rule, tag in rules:
try:
tags = self.dawg[rule]
except KeyError:
tags = []
self.dawg[rule] = tags
tags.append(tag)
self.tokens_range = tokens_range
def detect_negex_triggers(self, sentence: str):
# tokenize the sentence using a anti-whitespace pattern.
tokens = []
for match in _word_pattern.finditer(sentence):
tokens.append((match.start(), match.end()))
# use a DAWG matcher to locate all
matcher = self.dawg.matcher()
triggers = []
for i, (begin, end) in enumerate(tokens):
word = _not_word.sub('', sentence[begin:end].lower())
if len(word) > 0:
hits = matcher.advance(word)
for length, tags in hits:
first_token_idx = i + 1 - length
triggers.append(
(tokens[first_token_idx][0], tokens[i][1], tags))
return triggers
def test_matcher():
dawg = DAWG()
dawg[['a', 'b', 'c']] = True
dawg[['a', 'b', 'd']] = False
matcher = dawg.matcher()
assert matcher.advance('x') == []
assert matcher.advance('a') == []
assert matcher.advance('b') == []
assert matcher.advance('c') == [(3, True)]
assert matcher.advance('d') == []
class NegexTagger:
def __init__(self, rules: List[List[str]] = None):
if rules is None:
with (Path(__file__).parent / 'negex_triggers.txt').open('r') as f:
rules = make_rules(f)
self.dawg = DAWG()
for rule, tag in rules:
self.dawg[rule] = tag
def check_sentence(
self, sentence: str, terms: List[Tuple[int, int]]
) -> Tuple[List[Tuple[int, int]], List[Tuple[int, int]]]:
"""Checks the sentence for negated terms.
Args:
sentence (str): The sentence.
terms (~typing.List[~typing.Tuple[int, int]]):
A list of (start offset, end offset) tuples which indicate the locations of terms
within the sentence to test for negation.
Returns:
negated terms (~typing.List[~typing.Tuple[int, int]]):
The terms in the input which are negated. Start offset, end offset relative to the
sentence.
negation triggers (~typing.List[~typing.Tuple[int, int]]):
The spans of text which are negation triggers.
"""
if len(terms) == 0:
return [], []
# tokenize the sentence using a anti-whitespace pattern.
tokens = []
for match in _word_pattern.finditer(sentence):
tokens.append((match.start(), match.end()))
term_indices = []
for (term_start, term_end) in terms:
term_start_index = -1
for i, (token_start, token_end) in enumerate(tokens):
if term_start_index == -1 and token_start >= term_start:
term_start_index = i
if token_end >= term_end:
term_indices.append((term_start_index, i))
break
# use a DAWG matcher to locate all
matcher = self.dawg.matcher()
triggers = []
for i, (begin, end) in enumerate(tokens):
word = _not_word.sub('', sentence[begin:end].lower())
if len(word) > 0:
hits = matcher.advance(word)
for length, tag in hits:
first_token_idx = i + 1 - length
triggers.append((first_token_idx, i, tag))
negations = []
neg_triggers = []
for (term_start, term_end) in term_indices:
for i, (trigger_start, trigger_end, tag) in enumerate(triggers):
if term_start - trigger_end in range(1, 6):
if tag == 'PREN':
negations.append(
(tokens[term_start][0], tokens[term_end][1]))
neg_triggers.append(
(tokens[trigger_start][0], tokens[trigger_end][1]))
break
if trigger_start - term_end in range(1, 6) and tag == 'POST':
negations.append(
(tokens[term_start][0], tokens[term_end][1]))
neg_triggers.append(
(tokens[trigger_start][0], tokens[trigger_end][1]))
break
return negations, neg_triggers