def do_passives(self):
nodes = find_nodes(self,lambda n:n.features.get("Passive Voice",False))
for n in nodes:
curNeighbours = n.neighbors()
for subjNeigbour in multi_get(curNeighbours, subject_dependencies):
edge = (n,subjNeigbour)
self.del_edge(edge)
self.add_edge(edge,"obj")
for prepByNeigbour in multi_get(curNeighbours,["prep_by"]):
edge = (n,prepByNeigbour)
self.del_edge(edge)
self.add_edge(edge,"subj")
def do_acomp(self):
edges = find_edges(
self, lambda ((u, v)): self.edge_label(
(u, v)) == "acomp" and u.isPredicate)
for predNode, acompNode in edges:
neighbors = predNode.neighbors()
subjs = multi_get(neighbors, subject_dependencies)
if len(subjs) != 1:
# self.types.add("debug")
pass
else:
if (predNode.text[0].word
in self.modalVerbs) or (predNode.features.get(
"Lemma", "") in self.modalVerbs):
subj = subjs[0]
self.del_edge((predNode, acompNode))
self.add_edge((acompNode, subj), label=domain_label)
acompNode.isPredicate = True
self.del_edge((predNode, subj))
duplicate_all_incidents(gr=self,
source=predNode,
target=acompNode)
self.add_edge((acompNode, predNode), label=SOURCE_LABEL)
if (len(self.neighbors(predNode)) == 0) and (len(
predNode.text) == 1) and (predNode.text[0].word
in contractions):
self.del_node(predNode)
else:
self.types.add("acomp_as_modal")
else:
self.types.add("acomp_as_mwe")
merge_nodes(gr=self, node1=predNode, node2=acompNode)
def do_acomp(self):
edges = find_edges(self, lambda((u, v)):self.edge_label((u, v)) == "acomp" and u.isPredicate)
for predNode,acompNode in edges:
neighbors = predNode.neighbors()
subjs = multi_get(neighbors,subject_dependencies)
if len(subjs)!=1:
# self.types.add("debug")
pass
else:
if (predNode.text[0].word in self.modalVerbs) or (predNode.features.get("Lemma","") in self.modalVerbs):
subj = subjs[0]
self.del_edge((predNode,acompNode))
self.add_edge((acompNode,subj),label=domain_label)
acompNode.isPredicate=True
self.del_edge((predNode,subj))
duplicate_all_incidents(gr=self, source=predNode, target=acompNode)
self.add_edge((acompNode,predNode),label=SOURCE_LABEL)
if (len(self.neighbors(predNode))==0) and (len(predNode.text)==1) and (predNode.text[0].word in contractions):
self.del_node(predNode)
else:
self.types.add("acomp_as_modal")
else:
self.types.add("acomp_as_mwe")
merge_nodes(gr=self, node1=predNode, node2=acompNode)
def _fix(self):
# remove mark->that
edges = find_edges(self, lambda (u, v): self.edge_label(
(u, v)) == "mark")
for (u, v) in edges:
if (len(self.neighbors(v)) == 0) and (len(
v.text) == 1) and (v.text[0].word == "that"):
self.del_node(v)
return True
# rcmod with no relation to father
edges = find_edges(
self, lambda (u, v): (self.edge_label(
(u, v)) == "rcmod") and (not self.has_edge((v, u))))
for u, v in edges:
self.add_edge((v, u), label=ARG_LABEL)
return True
# prep collapse
edges = find_edges(
self, lambda (u, v): (self.edge_label((u, v)) == "prep") and
(len(self.neighbors(v)) == 1) and ("pobj" in v.neighbors()))
if edges:
for (u, v) in edges:
pobj = v.neighbors()["pobj"][0]
if not (self.has_edge((u, pobj))):
w = v.text[0]
u.surface_form += [w]
self.add_edge((u, pobj), label="prep_" + w.word)
self.del_node(v)
# fix dependency collapse bugs
edges = find_edges(
self, lambda (u, v): (self.edge_label(
(u, v)) == "pobj") and ("prep" not in u.incidents()))
for (u, v) in sorted(edges, key=lambda ((u, v)): u.minIndex()):
neighbors = u.neighbors()
candidates = [
n for n in multi_get(
neighbors,
[rel for rel in neighbors if rel.startswith("prepc_")])
if len(self.neighbors(n)) == 0
]
candidates.sort(key=lambda n: n.minIndex())
if len(candidates) > 0:
curToDel = candidates[0]
rel = self.edge_label((u, curToDel))
self.del_edge((u, v))
self.add_edge((u, v), label=rel)
self.del_node(curToDel)
def calcTopNodes(self):
all_accessible = multi_get(accessibility_wo_self(self), self.nodes())
for topNode in [n for n in self.nodes() if n not in all_accessible]:
topNode.makeTopNode()
change = True
while change:
change = False
for topNode in [n for n in self.nodes() if n.features.get("top", [])]:
for sourceNode in topNode.incidents().get(SOURCE_LABEL, []):
change = sourceNode.makeTopNode()
if topNode.isConj():
for n in self.neighbors(topNode):
change = n.makeTopNode()
def _fix(self):
# remove mark->that
edges = find_edges(self, lambda (u, v):self.edge_label((u, v)) == "mark")
for (u, v) in edges:
if (len(self.neighbors(v)) == 0) and (len(v.text) == 1) and (v.text[0].word == "that"):
self.del_node(v)
return True
# rcmod with no relation to father
edges = find_edges(self, lambda (u, v):(self.edge_label((u, v)) == "rcmod") and (not self.has_edge((v, u))))
for u, v in edges:
self.add_edge((v, u), label=ARG_LABEL)
return True
# prep collapse
edges = find_edges(self, lambda (u, v):(self.edge_label((u, v)) == "prep") and (len(self.neighbors(v)) == 1) and ("pobj" in v.neighbors()))
if edges:
for (u, v) in edges:
pobj = v.neighbors()["pobj"][0]
if not (self.has_edge((u, pobj))):
w = v.text[0]
u.surface_form += [w]
self.add_edge((u, pobj), label="prep_" + w.word)
self.del_node(v)
# fix dependency collapse bugs
edges = find_edges(self, lambda (u, v):(self.edge_label((u, v)) == "pobj") and ("prep" not in u.incidents()))
for (u, v) in sorted(edges,key=lambda((u,v)): u.minIndex()):
neighbors = u.neighbors()
candidates = [n for n in multi_get(neighbors, [rel for rel in neighbors if rel.startswith("prepc_")]) if len(self.neighbors(n)) == 0]
candidates.sort(key=lambda n:n.minIndex())
if len(candidates) > 0:
curToDel = candidates[0]
rel = self.edge_label((u, curToDel))
self.del_edge((u, v))
self.add_edge((u, v), label=rel)
self.del_node(curToDel)
def do_prop(self):
# prenominal of definite
edges = find_edges(self, lambda (u, v): self.edge_label(
(u, v)) == "amod")
for domain, mod in edges:
if domain.pos(
) in determined_labels: # the np by itself is definite
self.createPropRel(domain=domain, mod=mod)
mod.features["top"] = True
self.del_edge((domain, mod))
# copular on adjective or indefinite
# and sameAs otherwise
# find copular
nodes = find_nodes(
self, lambda n: len(n.text) == 1 and n.text[0].word in
copular_verbs and n.isPredicate)
for curNode in nodes:
curNeighbours = curNode.neighbors()
subjs = multi_get(curNeighbours, subject_dependencies)
objs = multi_get(curNeighbours, clausal_complements)
if not objs: objs = multi_get(curNeighbours, ["dep"])
others = [
n for n in self.neighbors(curNode) if n not in subjs + objs
]
if (len(objs) > 0) and (
len(subjs) > 0): #and (not others) and (len(objs) == 1):
others += objs[1:]
if others:
self.types.add("complicated BE")
obj = objs[0]
if len(objs) > 1:
self.types.add("debug")
for subj in subjs:
if 'Lemma' in curNode.features:
del (curNode.features['Lemma'])
if (subj in self.neighbors(obj)):
obj.features.update(curNode.features)
else:
if (not isDefinite(obj)) or (obj in curNeighbours.get(
"acomp", [])):
self.createPropRel(domain=subj, mod=obj)
head = obj
obj.surface_form += curNode.surface_form
else:
self.types.add("SameAs")
self.del_edge((curNode, subj))
if self.has_edge((curNode, obj)):
self.del_edge((curNode, obj))
# self.del_edges([(curNode, subj), (curNode, obj)])
copularNode = getCopular(self,
curNode.text[0].index,
features=curNode.features)
copularNode.surface_form = curNode.surface_form
self.add_edge((copularNode, subj),
label=FIRST_ENTITY_LABEL)
self.add_edge((copularNode, obj),
label=SECOND_ENTITY_LABEL)
head = copularNode
head.features.update(curNode.features)
for curFather in self.incidents(curNode):
if not self.has_edge((curFather, head)):
duplicateEdge(graph=self,
orig=(curFather, curNode),
new=(curFather, head))
for curOther in others:
if not self.has_edge((obj, curOther)):
duplicateEdge(graph=self,
orig=(curNode, curOther),
new=(head, curOther))
# erase "be" node
self.del_node(curNode)
# find appositions
for subj, obj in find_edges(
self, lambda edge: self.edge_label(edge) == "appos"):
# duplicate relations
for curFather in self.incidents(subj):
curIndex = curFather.features.get("apposIndex", 0) + 1
# curLabel = "{0},{1}".format(curIndex,self.edge_label((curFather,subj)))
curLabel = self.edge_label((curFather, subj))
self.del_edge((curFather, subj))
self.add_edge((curFather, subj), curLabel)
self.add_edge((curFather, obj), curLabel)
ls = curFather.features.get("dups", [])
ls.append((subj, obj))
curFather.features["dups"] = ls
curFather.features["apposIndex"] = curIndex
if (not isDefinite(subj)
and not isDefinite(obj)) or (obj in subj.neighbors().get(
"acomp", [])):
self.createPropRel(domain=subj, mod=obj)
obj.features["top"] = True
else:
# add new node
# TODO: subj here is a problem - should point to the comma or something
self.types.add("SameAs")
copularNode = getCopular(self, subj.text[0].index, features={})
copularNode.surface_form = []
self.add_edge((copularNode, subj), label=FIRST_ENTITY_LABEL)
self.add_edge((copularNode, obj), label=SECOND_ENTITY_LABEL)
self.del_edge((subj, obj))
def extract_entities(self):
ret = find_nodes(graph=self,
filterFunc=lambda node:
(not node.isPredicate) and multi_get(
node.incidents(), arguments_dependencies))
return ret
def do_prop(self):
# prenominal of definite
edges = find_edges(self, lambda (u, v):self.edge_label((u, v)) == "amod")
for domain, mod in edges:
if domain.pos() in determined_labels: # the np by itself is definite
self.createPropRel(domain=domain, mod=mod)
mod.features["top"] = True
self.del_edge((domain, mod))
# copular on adjective or indefinite
# and sameAs otherwise
# find copular
nodes = find_nodes(self, lambda n: len(n.text) == 1 and n.text[0].word in copular_verbs and n.isPredicate)
for curNode in nodes:
curNeighbours = curNode.neighbors()
subjs = multi_get(curNeighbours, subject_dependencies)
objs = multi_get(curNeighbours, clausal_complements)
if not objs: objs = multi_get(curNeighbours,["dep"])
others = [n for n in self.neighbors(curNode) if n not in subjs + objs]
if (len(objs)>0)and (len(subjs)>0): #and (not others) and (len(objs) == 1):
others+=objs[1:]
if others:
self.types.add("complicated BE")
obj = objs[0]
if len(objs)>1:
self.types.add("debug")
for subj in subjs:
if 'Lemma' in curNode.features: del(curNode.features['Lemma'])
if (subj in self.neighbors(obj)):
obj.features.update(curNode.features)
else:
if (not isDefinite(obj)) or (obj in curNeighbours.get("acomp", [])):
self.createPropRel(domain=subj, mod=obj)
head = obj
obj.surface_form += curNode.surface_form
else:
self.types.add("SameAs")
self.del_edge((curNode, subj))
if self.has_edge((curNode,obj)):
self.del_edge((curNode, obj))
# self.del_edges([(curNode, subj), (curNode, obj)])
copularNode = getCopular(self, curNode.text[0].index, features=curNode.features)
copularNode.surface_form = curNode.surface_form
self.add_edge((copularNode, subj),
label=FIRST_ENTITY_LABEL)
self.add_edge((copularNode, obj),
label=SECOND_ENTITY_LABEL)
head = copularNode
head.features.update(curNode.features)
for curFather in self.incidents(curNode):
if not self.has_edge((curFather, head)):
duplicateEdge(graph=self, orig=(curFather, curNode), new=(curFather, head))
for curOther in others:
if not self.has_edge((obj, curOther)):
duplicateEdge(graph=self, orig=(curNode, curOther), new=(head, curOther))
# erase "be" node
self.del_node(curNode)
# find appositions
for subj, obj in find_edges(self, lambda edge:self.edge_label(edge) == "appos"):
# duplicate relations
for curFather in self.incidents(subj):
curIndex = curFather.features.get("apposIndex", 0) + 1
# curLabel = "{0},{1}".format(curIndex,self.edge_label((curFather,subj)))
curLabel = self.edge_label((curFather, subj))
self.del_edge((curFather, subj))
self.add_edge((curFather, subj), curLabel)
self.add_edge((curFather, obj), curLabel)
ls = curFather.features.get("dups", [])
ls.append((subj, obj))
curFather.features["dups"] = ls
curFather.features["apposIndex"] = curIndex
if (not isDefinite(subj) and not isDefinite(obj)) or (obj in subj.neighbors().get("acomp", [])):
self.createPropRel(domain=subj, mod=obj)
obj.features["top"] = True
else:
# add new node
# TODO: subj here is a problem - should point to the comma or something
self.types.add("SameAs")
copularNode = getCopular(self, subj.text[0].index, features={})
copularNode.surface_form = []
self.add_edge((copularNode, subj),
label=FIRST_ENTITY_LABEL)
self.add_edge((copularNode, obj),
label=SECOND_ENTITY_LABEL)
self.del_edge((subj, obj))
def extract_entities(self):
ret = find_nodes(graph=self, filterFunc=lambda node: (not node.isPredicate) and multi_get(node.incidents(), arguments_dependencies))
return ret
def _fix(self):
# remove mark->that
edges = find_edges(
self, lambda u_v1: self.edge_label((u_v1[0], u_v1[1])) == "mark")
for (u, v) in edges:
if (len(self.neighbors(v)) == 0) and (len(
v.text) == 1) and (v.text[0].word == "that"):
self.del_node(v)
return True
# rcmod with no relation to father
edges = find_edges(
self, lambda u_v2: (self.edge_label(
(u_v2[0], u_v2[1])) == "rcmod") and (not self.has_edge(
(u_v2[1], u_v2[0]))))
for u, v in edges:
self.add_edge((v, u), label=ARG_LABEL)
return True
# prep collapse
edges = find_edges(
self, lambda u_v3: (self.edge_label(
(u_v3[0], u_v3[1])) == "prep") and
(len(self.neighbors(u_v3[1])) == 1) and
("pobj" in u_v3[1].neighbors()))
if edges:
for (u, v) in edges:
pobj = v.neighbors()["pobj"][0]
if not (self.has_edge((u, pobj))):
w = v.text[0]
u.surface_form += [w]
self.add_edge((u, pobj), label="prep_" + w.word)
self.del_node(v)
# fix dependency collapse bugs
edges = find_edges(
self, lambda u_v4: (self.edge_label(
(u_v4[0], u_v4[1])) == "pobj") and
("prep" not in u_v4[0].incidents()))
for (u, v) in sorted(edges, key=lambda u_v5: u_v5[0].minIndex()):
neighbors = u.neighbors()
candidates = [
n for n in multi_get(
neighbors,
[rel for rel in neighbors if rel.startswith("prepc_")])
if len(self.neighbors(n)) == 0
]
candidates.sort(key=lambda n: n.minIndex())
if len(candidates) > 0:
curToDel = candidates[0]
rel = self.edge_label((u, curToDel))
self.del_edge((u, v))
self.add_edge((u, v), label=rel)
self.del_node(curToDel)
# change agent edges with "prep_by"
edges = find_edges(self, lambda edge:
(self.edge_label(edge) == "agent"))
for edge in edges:
self.del_edge(edge)
self.add_edge(edge, label="prep_by")
# #add xcomp inverse node
# edges = find_edges(self, lambda (u,v):self.edge_label((u,v)) == "xcomp" and u.isPredicate and v.isPredicate)
# for (u,v) in edges:
# if not self.has_edge((v, u)):
# self.add_edge((v,u), label=SOURCE_LABEL)
# self.types.add("infinitives")
# return True
# if not multi_get(v.neighbors(),subject_dependencies):
# rcmodParentIncidents = u.incidents().get("rcmod",[])
# if len(rcmodParentIncidents)==1:
# subj = rcmodParentIncidents[0]
# if not self.has_edge((v,subj)):
# self.add_edge((v,subj),label=ARG_LABEL)
return False
