def getCoreferent(self, role):
coreferent_phrase = None
# for coord purposes
coreferent_clause = None
# get number of phrase - if coreferent is coordination, it will be choosen based on number
phrase_number = role.phrase.getNumberCategory() if role.phrase != None else 0
# sequentially add clauses to the list order
clauses = []
clause_found = False
for sentence in self.sentences:
for clause in sentence.clauses:
# found containing clause
if clause == role.getRelation().containing_clause:
clause_found = True
# if clause was just found and the antecedent is not a pronoun, search also current clause
if role.phrase == None or (role.phrase != None and not 'k3yR' in role.phrase.tokens[0].tag):
clauses.insert(0, clause)
# clause wasn't found yet, add clause at the beginning
elif not clause_found:
clauses.insert(0, clause)
# clause was found, add clause at the end
else:
clauses.append(clause)
# find coreferent
i = 0
while i < len(clauses) and coreferent_phrase == None:
j = 0
while j < len(clauses[i].phrases) and coreferent_phrase == None:
phrase_role = clauses[i].phrases[j].hasRole(role.second_level_role)
# newer version - search also base roles
if not phrase_role:
phrase_role = clauses[i].phrases[j].hasBaseRole(role.second_level_role)
# update role
if phrase_role:
phrase_role.second_level_role = role.second_level_role
if phrase_role and phrase_role.filledWithNE():
coreferent_phrase = clauses[i].phrases[j]
coreferent_clause = clauses[i]
j += 1
i += 1
# if phrase is coordination and antecedent wants just one entity
if phrase_number == 1 and isinstance(coreferent_phrase, NPhrase) and coreferent_phrase.is_coordination:
sub_phrases = coreferent_clause.getDependentPhrases(coreferent_phrase)
new_coreferent = None
i = len(sub_phrases) - 1
# find the latest NE in given coordination
while i >= 0 and new_coreferent == None:
if Utils.isNamedEntity(sub_phrases[i]):
new_coreferent = sub_phrases[i]
i = i - 1
if new_coreferent != None:
coreferent_phrase = new_coreferent
return coreferent_phrase
def getStocks(self, relation):
stocks = []
stocks_phrases = relation.getSecondLevelRoles('<actor_stock:1>')
# collect all stock phrases from given relation
for candidate in stocks_phrases:
if candidate.coreferent != None and Utils.isNamedEntity(candidate.coreferent) and not candidate.coreferent in stocks_phrases:
# get named entities from given phrase
for entity_str in Utils.getNamedEntities(candidate.coreferent):
if not entity_str.replace('_', ' ') in stocks:
stocks.append(entity_str.replace('_', ' '))
return stocks
def getAgencies(self, relation):
agencies = []
agencies_phrases = relation.getSecondLevelRoles('<actor_agency:1>')
# collect all agency phrases from given relation
for candidate in agencies_phrases:
if candidate.coreferent != None and Utils.isNamedEntity(candidate.coreferent) and not candidate.coreferent in agencies_phrases:
# get named entities from given phrase
for entity_str in Utils.getNamedEntities(candidate.coreferent):
# is it relevant agency string? Dfens against bugs and wrong parses
if RoleResolver.isRelevantAgencyEntity(entity_str) and not entity_str.replace('_', ' ') in agencies:
agencies.append(entity_str.replace('_', ' '))
return agencies
def filledWithNE(self):
return Utils.isNamedEntity(self.phrase) if self.phrase != None else False
def matchesPhrase(self, phrase):
return Utils.isNamedEntity(phrase)