def build_tree(text):
""" Build tree from *.dis file
:type text: string
:param text: RST tree read from a *.dis file
"""
tokens = text.strip().replace('//TT_ERR', '').replace(
'\n', '').replace('(', ' ( ').replace(')', ' ) ').split()
queue = RstTree.process_text(tokens)
stack = []
while queue:
token = queue.pop(0)
if token == ')':
# If ')', start processing
content = [] # Content in the stack
while stack:
cont = stack.pop()
if cont == '(':
break
else:
content.append(cont)
content.reverse() # Reverse to the original (stack) order
# Parse according to the first content word
if len(content) < 2:
raise ValueError("content = {}".format(content))
label = content.pop(0)
if label in ['Root', 'Nucleus', 'Satellite']:
node = SpanNode(prop=label)
node.create_node(content)
stack.append(node)
elif label == 'span':
# Merge
beginindex = int(content.pop(0))
endindex = int(content.pop(0))
stack.append(('span', beginindex, endindex))
elif label == 'leaf':
# Merge
eduindex = int(content.pop(0))
RstTree.check_content(label, content)
stack.append(('leaf', eduindex, eduindex))
elif label == 'rel2par':
# Merge
relation = content.pop(0)
RstTree.check_content(label, content)
stack.append(('relation', relation))
elif label == 'text':
# Merge
txt = RstTree.create_text(content)
stack.append(('text', txt))
else:
raise ValueError(
"Unrecognized parsing label: {} \n\twith content = {}\n\tstack={}\n\tqueue={}"
.format(label, content, stack, queue))
else:
# else, keep push into the stack
stack.append(token)
return stack[-1]
def init(self, doc):
""" Using text to initialize Queue
:type doc: Doc instance
:param doc:
"""
N = len(doc.edu_dict)
for idx in range(1, N + 1, 1):
node = SpanNode(prop=None)
node.text = doc.edu_dict[idx]
node.edu_span, node.nuc_span = (idx, idx), (idx, idx)
node.nuc_edu = idx
self.Queue.append(node)
def binarize_tree(tree):
""" Convert a general RST tree to a binary RST tree
:type tree: instance of SpanNode
:param tree: a general RST tree
"""
queue = [tree]
while queue:
node = queue.pop(0)
queue += node.nodelist
# Construct binary tree
if len(node.nodelist) == 2:
node.lnode = node.nodelist[0]
node.rnode = node.nodelist[1]
# Parent node
node.lnode.pnode = node
node.rnode.pnode = node
elif len(node.nodelist) > 2:
# Remove one node from the nodelist
lc = node.nodelist[0].prop
mark = 1
for nl in node.nodelist:
mark &= nl.visited
# if not mark:
# # if not [0^nl.visited for nl in node.nodelist]:
# if node.relation:
# print(RstTree.extract_relation(node.relation))
# print([RstTree.extract_relation(l.relation) for l in node.nodelist])
if len(set([l.prop for l in node.nodelist])) == 1:
node.visited = True
# for nl in node.nodelist:
# nl.visited = True
node.lnode = node.nodelist.pop(0)
newnode = SpanNode(node.nodelist[0].prop)
newnode.nodelist += node.nodelist
# Right-branching
node.rnode = newnode
# Parent node
node.lnode.pnode = node
node.rnode.pnode = node
if node.visited:
newnode.visited = True
queue.insert(0, newnode)
# Clear nodelist for the current node
node.nodelist = []
return tree
def init(self, doc):
""" Using text to initialize Queue
:type doc: Doc instance
:param doc:
"""
if not isinstance(doc, Doc):
raise ValueError("doc should be an instance of Doc")
N = len(doc.edu_dict)
for idx in range(1, N + 1, 1):
node = SpanNode(prop=None)
node.text = doc.edu_dict[idx]
node.edu_span, node.nuc_span = (idx, idx), (idx, idx)
node.nuc_edu = idx
self.Queue.append(node)
def flat_tree(tree):
queue = [tree]
while queue:
node = queue.pop(0)
queue += node.nodelist
# Construct binary tree
if len(node.nodelist) == 2:
node.lnode = node.nodelist[0]
node.rnode = node.nodelist[1]
# Parent node
node.lnode.pnode = node
node.rnode.pnode = node
elif len(node.nodelist) > 2:
if len(set([l.prop for l in node.nodelist])) != 1:
node.lnode = node.nodelist.pop(0)
newnode = SpanNode(node.nodelist[0].prop)
newnode.nodelist += node.nodelist
# Right-branching
node.rnode = newnode
# Parent node
node.lnode.pnode = node
node.rnode.pnode = node
queue.insert(0, newnode)
# reset nodelist for the current node
node.nodelist = [node.lnode, node.rnode]
return tree
def binarize_tree(tree):
""" Convert a general RST tree to a binary RST tree
:type tree: instance of SpanNode
:param tree: a general RST tree
"""
queue = [tree]
while queue:
node = queue.pop(0)
queue += node.nodelist
# Construct binary tree
if len(node.nodelist) == 2:
node.lnode = node.nodelist[0]
node.rnode = node.nodelist[1]
# Parent node
node.lnode.pnode = node
node.rnode.pnode = node
elif len(node.nodelist) > 2:
# Remove one node from the nodelist
node.lnode = node.nodelist.pop(0)
newnode = SpanNode(node.nodelist[0].prop)
newnode.nodelist += node.nodelist
# Right-branching
node.rnode = newnode
# Parent node
node.lnode.pnode = node
node.rnode.pnode = node
# Add to the head of the queue
# So the code will keep branching
# until the nodelist size is 2
queue.insert(0, newnode)
# Clear nodelist for the current node
node.nodelist = []
return tree
def operate(self, action_tuple):
""" According to parsing label to modify the status of
the Stack/Queue
"""
action, form = action_tuple
if action == 'Shift':
if len(self.Queue) == 0:
raise ActionError("Shift action error")
node = self.Queue.pop(0)
self.Stack.append(node)
elif action == 'Reduce':
if len(self.Stack) < 2:
raise ActionError("Reduce action error")
rnode = self.Stack.pop()
lnode = self.Stack.pop()
# Create a new node
# Assign a value to prop, only when it is someone's
# children node
node = SpanNode(prop=None)
# Children node
node.lnode, node.rnode = lnode, rnode
# Parent node of children nodes
node.lnode.pnode, node.rnode.pnode = node, node
# Node text: concatenate two word lists
node.text = lnode.text + rnode.text
# EDU span
node.edu_span = (lnode.edu_span[0], rnode.edu_span[1])
# Nuc span / Nuc EDU
node.form = form
if form == 'NN':
node.nuc_edu = lnode.nuc_edu
node.lnode.prop = "Nucleus"
node.rnode.prop = "Nucleus"
elif form == 'NS':
node.nuc_edu = lnode.nuc_edu
node.lnode.prop = "Nucleus"
node.rnode.prop = "Satellite"
elif form == 'SN':
node.nuc_edu = rnode.nuc_edu
node.lnode.prop = "Satellite"
node.rnode.prop = "Nucleus"
else:
raise ValueError("Unrecognized form: {}".format(form))
self.Stack.append(node)
else:
raise ValueError("Unrecognized parsing action: {}".format(action))
def operate(self, action_tuple):
""" According to parsing label to modify the status of
the Stack/Queue
"""
action, form = action_tuple
if action == 'Shift':
if len(self.Queue) == 0:
raise ActionError("Shift action error")
node = self.Queue.pop(0)
self.Stack.append(node)
elif action == 'Reduce':
if len(self.Stack) < 2:
raise ActionError("Reduce action error")
rnode = self.Stack.pop()
lnode = self.Stack.pop()
# Create a new node
# Assign a value to prop, only when it is someone's
# children node
node = SpanNode(prop=None)
# Children node
node.lnode, node.rnode = lnode, rnode
# dependency
l = lnode.dependency
r = rnode.dependency
node.dependency = np.average([l, r], axis=0)
# Parent node of children nodes
node.lnode.pnode, node.rnode.pnode = node, node
node.nodelist = [node.lnode, node.rnode]
node.lnode.pnode, node.rnode.pnode = node, node
# Node text: concatenate two word lists
node.text = lnode.text + rnode.text
# EDU span
node.edu_span = (lnode.edu_span[0], rnode.edu_span[1])
# Nuc span / Nuc EDU
node.form = form
if form == 'NN':
node.nuc_span = (lnode.edu_span[0], rnode.edu_span[1])
node.nuc_edu = lnode.nuc_edu
node.lnode.prop = "Nucleus"
node.rnode.prop = "Nucleus"
elif form == 'N~':
node.nuc_span = (lnode.edu_span[0], rnode.edu_span[1])
node.nuc_edu = lnode.nuc_edu
node.lnode.prop = "Nucleus"
node.rnode.prop = "Nucleus"
elif form == 'NS':
node.nuc_span = lnode.edu_span
node.nuc_edu = lnode.nuc_edu
node.lnode.prop = "Nucleus"
node.rnode.prop = "Satellite"
elif form == 'SN':
node.nuc_span = rnode.edu_span
node.nuc_edu = rnode.nuc_edu
node.lnode.prop = "Satellite"
node.rnode.prop = "Nucleus"
else:
raise ValueError("Unrecognized form: {}".format(form))
self.Stack.append(node)
elif action == 'R~':
if len(self.Stack) < 2:
raise ActionError("Reduce action error")
enode = self.Stack.pop()
snode = self.Stack.pop()
node = SpanNode(prop=None)
if enode.form == 'N~':
node.nodelist.append(snode)
node.nodelist.extend(enode.nodelist)
else:
node.nodelist = [snode, enode]
# Children node
node.lnode, node.rnode = node.nodelist[0], node.nodelist[-1]
# dependency
node.dependency = np.average([n.dependency for n in node.nodelist],
axis=0)
# Parent node of children nodes
node.lnode.pnode, node.rnode.pnode = node, node
lnode, rnode = node.lnode, node.rnode
# Node text: concatenate two word lists
node.text = []
for nl in node.nodelist:
node.text += nl.text
# EDU span
node.edu_span = (lnode.edu_span[0], rnode.edu_span[1])
node.form = 'N~'
node.nuc_span = (lnode.edu_span[0], rnode.edu_span[1])
node.nuc_edu = lnode.nuc_edu
node.lnode.prop = "Nucleus"
node.rnode.prop = "Nucleus"
self.Stack.append(node)
else:
raise ValueError("Unrecognized parsing action: {}".format(action))