def revise_node(content, amr_nodes_content, amr_nodes_acronym):
'''
In case of single '()' contains multiple nodes
e.x. (m / moment :poss p5)
:param str context:
:param dict amr_nodes_content: content as key
:param dict amr_nodes_acronym: acronym as key
'''
m = re.search('\w+\s/\s\S+\s+(.+)', content.replace('\n', ''))
if m and ' / name' not in content and ':polarity -' not in content:
arg_nodes = []
acr = re.search('\w+\s/\s\S+', content).group().split(' / ')[0]
nodes = re.findall('\S+\s\".+\"|\S+\s\S+', m.group(1))
for i in nodes:
i = re.search('(:\S+)\s(.+)', i)
role = i.group(1)
concept = i.group(2).strip(')')
if concept in amr_nodes_acronym:
node = copy.copy(amr_nodes_acronym[concept])
node.next_nodes = []
else: # in case of (d / date-entity :year 2012)
node = Node(name=concept)
amr_nodes_acronym[concept] = node
node.edge_label = role
arg_nodes.append(node)
amr_nodes_acronym[acr].next_nodes = arg_nodes
amr_nodes_content[content].next_nodes = arg_nodes
def generate_nodes_multiple(content, amr_nodes_content, amr_nodes_acronym):
'''
Generate Node object for nested '()'
:param str context:
:param dict amr_nodes_content: content as key
:param dict amr_nodes_acronym: acronym as key
'''
try:
assert content.count('(') > 1 and content.count(')') > 1
assert content.count('(') == content.count(')')
except AssertionError:
raise Exception('Unmatched parenthesis')
#note that if we want to get complete content, we need to complete subgraphs in amr_nodes_content, however if we really complete it, the following can't be indexed in amr_nodes_content
_content = content #because content will gradually move its components, so use _content to back up for amr_nodes_content
org = content #original_content difference between above is It don't remove :name
arg_nodes = []
is_named_entity = False
# Remove existing nodes from the content, and link these nodes to the root
# of the subtree
for i in sorted(amr_nodes_content, key=len, reverse=True):
if i in content:
e = content.find(i)
s = content[:e].rfind(':')
role = re.search(':\S+\s', content[s:e]).group() # Edge label
amr_nodes_content[i].edge_label = role.strip()
if ':name' in role:
is_named_entity = True
ne = amr_nodes_content[i]
else:
arg_nodes.append(amr_nodes_content[i])
if ':name' not in role:
org = org.replace(role + i, '', 1)
content = content.replace(role + i, '', 1)
predict_event = re.search('\w+\s/\s\S+', content).group().split(' / ')
if predict_event:
acr = predict_event[0] # Acronym
ful = predict_event[1] # Full name
else:
acr, ful = '-', '-'
# In case of :polarity -
is_polarity = True if re.search(":polarity\s-", content) else False
nodes = re.findall(':\S+\s\S+', content)
for i in nodes:
i = re.search('(:\S+)\s(\S+)', i)
role = i.group(1)
concept = i.group(2).strip("()")
if role == ':wiki' and is_named_entity:
continue
if role in [':polarity', ':quant', ':age', ':value']:
continue
if concept in amr_nodes_acronym:
node = copy.copy(amr_nodes_acronym[concept])
content = content.replace(i.group(0), "")
# In case of (d / date-entity :year 2012)
else:
node = Node(name=concept)
amr_nodes_acronym[concept] = node
# if re.search("\s+"+concept+"[^\d]", content, ):
# content.replace(concept, )
node.edge_label = role
arg_nodes.append(node)
# Named entity is a special node, so the subtree of a
# named entity will be merged. For example,
# (p / person :wiki -
# :name (n / name
# :op1 "Pascale"))
# will be merged as one node.
# According to AMR Specification, "we fill the :instance
# slot from a special list of standard AMR named entity types".
# Thus, for named entity node, we will use entity type
# (p / person in the example above) instead of :instance
if is_named_entity:
# Get Wikipedia title:
if re.match('.+:wiki\s-.*', content):
wikititle = '-' # Entity is NIL, Wiki title does not exist
else:
m = re.search(':wiki\s\"(.+?)\"', content)
if m:
wikititle = urllib.parse.unquote_plus(m.group(1)) # Wiki title
else:
wikititle = '' # There is no Wiki title information
new_node = Node(name=acr,
ful_name=ful,
next_nodes=arg_nodes,
parents=set(),
edge_label=ne.ful_name,
is_entity=True,
entity_type=ful,
entity_name=ne.entity_name,
wiki=wikititle,
polarity=is_polarity,
content=content,
original_content=org)
amr_nodes_content[_content] = new_node
amr_nodes_acronym[acr] = new_node
elif len(arg_nodes) > 0:
new_node = Node(name=acr,
ful_name=ful,
next_nodes=arg_nodes,
parents=set(),
polarity=is_polarity,
content=content,
original_content=_content)
amr_nodes_content[_content] = new_node
amr_nodes_acronym[acr] = new_node
for child in new_node.next_nodes:
child.parents.add(new_node)
def generate_node_single(content, amr_nodes_content, amr_nodes_acronym):
'''
Generate Node object for single '()'
:param str context:
:param dict amr_nodes_content: content as key
:param dict amr_nodes_acronym: acronym as key
'''
is_named_entity = False
try:
assert content.count('(') == 1 and content.count(')') == 1
except AssertionError:
raise Exception('Unmatched parenthesis')
predict_event = re.search('(\w+)\s/\s(\S+)', content)
if predict_event:
acr = predict_event.group(1) # Acronym
ful = predict_event.group(2).strip(')') # Full name
else:
acr, ful = '-', '-'
# In case of :polarity -
is_polarity = True if re.search(":polarity\s-", content) else False
# :ARG ndoes
arg_nodes = []
nodes = re.findall(':\S+\s\S+', content)
for i in nodes:
i = re.search('(:\S+)\s(\S+)', i)
role = i.group(1)
concept = i.group(2).strip(')')
if role == ':wiki' and is_named_entity:
continue
if role == ':polarity':
continue
if concept in amr_nodes_acronym:
node = copy.copy(amr_nodes_acronym[concept])
node.next_nodes = []
# In case of (d / date-entity :year 2012)
else:
node = Node(name=concept, original_content=concept)
amr_nodes_acronym[concept] = node
node.edge_label = role
arg_nodes.append(node)
# Node is a named entity
names = re.findall(':op\d\s\"\S+\"', content)
if len(names) > 0:
entity_name = ''
for i in names:
entity_name += re.match(':op\d\s\"(\S+)\"', i).group(1) + ' '
entity_name = urllib.parse.unquote_plus(entity_name.strip())
new_node = Node(name=acr,
ful_name=ful,
next_nodes=arg_nodes,
parents=set(),
entity_name=entity_name,
polarity=is_polarity,
content=content,
original_content=content)
amr_nodes_content[content] = new_node
amr_nodes_acronym[acr] = new_node
else:
new_node = Node(name=acr,
ful_name=ful,
next_nodes=arg_nodes,
parents=set(),
polarity=is_polarity,
content=content,
original_content=content)
amr_nodes_content[content] = new_node
amr_nodes_acronym[acr] = new_node
def generate_nodes_multiple(content, amr_nodes_content, amr_nodes_acronym):
'''
Generate Node object for nested '()'
:param str context:
:param dict amr_nodes_content: content as key
:param dict amr_nodes_acronym: acronym as key
'''
try:
assert content.count('(') > 1 and content.count(')') > 1
assert content.count('(') == content.count(')')
except AssertionError:
raise Exception('Unmatched parenthesis')
_content = content
arg_nodes = []
is_named_entity = False
# Remove existing nodes from the content, and link these nodes to the root
# of the subtree
for i in sorted(amr_nodes_content, key=len, reverse=True):
if i in content:
e = content.find(i)
s = content[:e].rfind(':')
role = re.search(':\S+\s', content[s:e]).group() # Edge label
content = content.replace(role+i, '', 1)
amr_nodes_content[i].edge_label = role.strip()
if ':name' in role:
is_named_entity = True
ne = amr_nodes_content[i]
else:
arg_nodes.append(amr_nodes_content[i])
predict_event = re.search('\w+\s/\s\S+', content).group().split(' / ')
if predict_event:
acr = predict_event[0] # Acronym
ful = predict_event[1] # Full name
else:
acr, ful = '-', '-'
# In case of :polarity -
is_polarity = True if re.search(":polarity\s-", content) else False
nodes = re.findall(':\S+\s\S+', content)
for i in nodes:
i = re.search('(:\S+)\s(\S+)', i)
role = i.group(1)
concept = i.group(2).strip(')')
if role == ':wiki' and is_named_entity:
continue
if role == ':polarity':
continue
if concept in amr_nodes_acronym:
node = copy.copy(amr_nodes_acronym[concept])
node.next_nodes = []
# In case of (d / date-entity :year 2012)
else:
node = Node(name=concept)
amr_nodes_acronym[concept] = node
node.edge_label = role
arg_nodes.append(node)
# Named entity is a special node, so the subtree of a
# named entity will be merged. For example,
# (p / person :wiki -
# :name (n / name
# :op1 "Pascale"))
# will be merged as one node.
# According to AMR Specification, "we fill the :instance
# slot from a special list of standard AMR named entity types".
# Thus, for named entity node, we will use entity type
# (p / person in the example above) instead of :instance
if is_named_entity:
# Get Wikipedia title:
if re.match('.+:wiki\s-.*', content):
wikititle = '-' # Entity is NIL, Wiki title does not exist
else:
m = re.search(':wiki\s\"(.+?)\"', content)
if m:
wikititle = urllib.parse.unquote_plus(m.group(1)) # Wiki title
else:
wikititle = '' # There is no Wiki title information
new_node = Node(name=acr, ful_name=ful, next_nodes=arg_nodes,
edge_label=ne.ful_name, is_entity=True,
entity_type=ful, entity_name=ne.entity_name,
wiki=wikititle, polarity=is_polarity, content=content)
amr_nodes_content[_content] = new_node
amr_nodes_acronym[acr] = new_node
elif len(arg_nodes) > 0:
new_node = Node(name=acr, ful_name=ful, next_nodes=arg_nodes,
polarity=is_polarity, content=content)
amr_nodes_content[_content] = new_node
amr_nodes_acronym[acr] = new_node