def parse_semantic_dependencies(self, tokenized_sentence, tags, lemmas,
heads, deprels, language):
worker = self.get_worker(language)
sent = NLPSentence()
sent['words'] = tokenized_sentence
sent['tags'] = tags
sent['lemmas'] = lemmas
sent['heads'] = heads
# Convert from 1-based indexing for back-compatibility.
#sent['heads'] = [h-1 for h in heads]
sent['dependency_relations'] = deprels
sent.compute_semantic_dependencies(worker)
num_predicates = len(sent['predicate_names'])
predicates = ['_' for token in tokenized_sentence]
argument_lists = [['_' for k in xrange(num_predicates)] \
for token in tokenized_sentence]
for k in xrange(num_predicates):
name = sent['predicate_names'][k]
p = sent['predicate_indices'][k]
predicates[p] = name
for l in xrange(len(sent['argument_roles'][k])):
role = sent['argument_roles'][k][l]
a = sent['argument_indices'][k][l]
argument_lists[a][k] = role
return predicates, argument_lists
def parse_semantic_dependencies(self, tokenized_sentence, tags, lemmas,
heads, deprels, language):
worker = self.get_worker(language)
sent = NLPSentence()
sent['words'] = tokenized_sentence
sent['tags'] = tags
sent['lemmas'] = lemmas
sent['heads'] = heads
# Convert from 1-based indexing for back-compatibility.
#sent['heads'] = [h-1 for h in heads]
sent['dependency_relations'] = deprels
sent.compute_semantic_dependencies(worker)
num_predicates = len(sent['predicate_names'])
predicates = ['_' for token in tokenized_sentence]
argument_lists = [['_' for k in xrange(num_predicates)] \
for token in tokenized_sentence]
for k in xrange(num_predicates):
name = sent['predicate_names'][k]
p = sent['predicate_indices'][k]
predicates[p] = name
for l in xrange(len(sent['argument_roles'][k])):
role = sent['argument_roles'][k][l]
a = sent['argument_indices'][k][l]
argument_lists[a][k] = role
return predicates, argument_lists
def recognize_entities(self, tokenized_sentence, tags, language):
worker = self.get_worker(language)
sent = NLPSentence()
sent['words'] = tokenized_sentence
sent['tags'] = tags
sent.compute_entities(worker)
entity_tags = sent['entity_tags']
return entity_tags
def recognize_entities(self, tokenized_sentence, tags, language):
worker = self.get_worker(language)
sent = NLPSentence()
sent['words'] = tokenized_sentence
sent['tags'] = tags
sent.compute_entities(worker)
entity_tags = sent['entity_tags']
return entity_tags
def parse(self, tokenized_sentence, tags, lemmas, language):
worker = self.get_worker(language)
sent = NLPSentence()
sent['words'] = tokenized_sentence
sent['tags'] = tags
sent['lemmas'] = lemmas
sent.compute_syntactic_dependencies(worker)
# Convert to 1-based indexing for back-compatibility.
#heads = [h+1 for h in sent['heads']]
heads = sent['heads']
deprels = sent['dependency_relations']
return heads, deprels
def parse(self, tokenized_sentence, tags, lemmas, language):
worker = self.get_worker(language)
sent = NLPSentence()
sent['words'] = tokenized_sentence
sent['tags'] = tags
sent['lemmas'] = lemmas
sent.compute_syntactic_dependencies(worker)
# Convert to 1-based indexing for back-compatibility.
#heads = [h+1 for h in sent['heads']]
heads = sent['heads']
deprels = sent['dependency_relations']
return heads, deprels
def tag(self, tokenized_sentence, language):
worker = self.get_worker(language)
sent = NLPSentence()
sent['words'] = tokenized_sentence
sent.compute_morphology(worker)
tags = sent['tags']
if sent['lemmas'] != None:
lemmas = sent['lemmas']
else:
lemmas = ['_' for token in tokenized_sentence]
if sent['morphological_tags'] != None:
feats = ['|'.join(morph) if len(morph) > 0 else '_' \
for morph in sent['morphological_tags']]
else:
feats = ['_' for token in tokenized_sentence]
return tags, lemmas, feats
def resolve_coreferences(self, all_tokenized_sentences, all_tags,
all_lemmas, all_heads, all_deprels,
all_entity_tags, language):
worker = self.get_worker(language)
sents = []
for j, tokenized_sentence in enumerate(all_tokenized_sentences):
sent = NLPSentence()
sent['words'] = tokenized_sentence
sent['tags'] = all_tags[j]
sent['lemmas'] = all_lemmas[j]
sent['heads'] = all_heads[j]
# Convert from 1-based indexing for back-compatibility.
#sent['heads'] = [h-1 for h in all_heads[j]]
sent['dependency_relations'] = all_deprels[j]
# For now, don't use this (must be coded as spans).
sent['entity_tags'] = all_entity_tags[j]
sents.append(sent)
doc = NLPDocument(sents)
doc.compute_coreferences(worker)
# Convert from spans to coref info.
all_coref_info = []
for sent in doc['sentences']:
spans = []
for (start, end, name) in sent['coreference_spans']:
span = Span(start, end, name)
spans.append(span)
coref_info = nlp_utils.construct_coreference_info_from_spans( \
spans, len(sent['words']))
all_coref_info.append(coref_info)
return all_coref_info
def tag(self, tokenized_sentence, language):
worker = self.get_worker(language)
sent = NLPSentence()
sent['words'] = tokenized_sentence
sent.compute_morphology(worker)
tags = sent['tags']
if sent['lemmas'] != None:
lemmas = sent['lemmas']
else:
lemmas = ['_' for token in tokenized_sentence]
if sent['morphological_tags'] != None:
feats = ['|'.join(morph) if len(morph) > 0 else '_' \
for morph in sent['morphological_tags']]
else:
feats = ['_' for token in tokenized_sentence]
return tags, lemmas, feats
