def fall_back_left_branching_token(clean_tokens):
tree = HybridTree()
for i, token in enumerate(clean_tokens):
token.set_edge_label('_')
tree.add_node(i, token, True)
if i == 0:
tree.add_to_root(i)
else:
tree.add_child(i - 1, i)
return tree
def fall_back_left_branching(forms, poss):
tree = HybridTree()
for i, (form, pos) in enumerate(zip(forms, poss)):
token = construct_conll_token(form, pos)
token.set_edge_label('_')
tree.add_node(i, token, True)
if i == 0:
tree.add_to_root(i)
else:
tree.add_child(i - 1, i)
return tree
def hybrid_tree_1():
tree = HybridTree()
tree.add_node('v1', CoNLLToken('Piet', '_', 'NP', 'NP', '_', 'SBJ'), True)
tree.add_node('v21', CoNLLToken('Marie', '_', 'N', 'N', '_', 'OBJ'), True)
tree.add_node('v', CoNLLToken('helpen', '_', 'V', 'V', '_', 'ROOT'), True)
tree.add_node('v2', CoNLLToken('lezen', '_', 'V', 'V', '_', 'VBI'), True)
tree.add_child('v', 'v2')
tree.add_child('v', 'v1')
tree.add_child('v2', 'v21')
tree.add_to_root('v')
tree.reorder()
return tree
def multi_const_tree():
tree = HybridTree("multi")
tree.add_node('1.1', ConstituentTerminal('A', 'pA'), True, True)
tree.add_node('2.1', ConstituentTerminal('B', 'pB'), True, True)
tree.add_node('1.2', ConstituentTerminal('C', 'pC'), True, True)
tree.add_node('2.2', ConstituentTerminal('D', 'pD'), True, True)
tree.add_node('1', ConstituentCategory('E'), False, True)
tree.add_node('2', ConstituentCategory('F'), False, True)
for p in ['2', '1']:
tree.add_to_root(p)
for c in ['1', '2']:
tree.add_child(p, p + '.' + c)
return tree
def multi_dep_tree():
tree = HybridTree('multi')
tree.add_node('1', CoNLLToken('A', '_', 'pA', 'pA', '_', 'dA'), True)
tree.add_node('211', CoNLLToken('B', '_', 'pB', 'pB', '_', 'dB'), True)
tree.add_node('11', CoNLLToken('C', '_', 'pC', 'pC', '_', 'dC'), True)
tree.add_node('2', CoNLLToken('D', '_', 'pD', 'pD', '_', 'dD'), True)
tree.add_node('21', CoNLLToken('E', '_', 'pE', 'pE', '_', 'dE'), True)
tree.add_to_root('2')
tree.add_to_root('1')
for c in ['21', '211']:
tree.add_child('2', c)
tree.add_child('1', '11')
tree.reorder()
return tree
def disconnect_punctuation(trees):
"""
:param trees: corpus of hybrid trees
:type trees: __generator[HybridTree]
:return: corpus of hybrid trees
:rtype: __generator[GeneralHybridTree]
lazily disconnect punctuation from each hybrid tree in a corpus of hybrid trees
"""
for tree in trees:
tree2 = HybridTree(tree.sent_label())
for root_id in tree.root:
if not is_punctuation(tree.node_token(root_id).form()):
tree2.add_to_root(root_id)
for id in tree.full_yield():
token = tree.node_token(id)
if not is_punctuation(token.form()):
parent = tree.parent(id)
while parent and parent not in tree.root and is_punctuation(
tree.node_token(parent).form()):
parent = tree.parent(parent)
if parent and is_punctuation(tree.node_token(parent).form()):
tree2.add_to_root(id)
else:
tree2.add_child(parent, id)
tree2.add_node(id, token, True, True)
else:
tree2.add_node(id, token, True, False)
if tree2:
# basic sanity checks
if not tree2.root \
and len(tree2.id_yield()) == 0 \
and len(tree2.nodes()) == len(tree2.full_yield()):
# Tree consists only of punctuation
continue
elif not tree2.root \
or tree2.n_nodes() != len(tree2.id_yield()) \
or len(tree2.nodes()) != len(tree2.full_yield()):
print(tree)
print(tree2)
print(tree2.sent_label())
print("Root:", tree2.root)
print("Nodes: ", tree2.n_nodes())
print("Id_yield:", len(tree2.id_yield()), tree2.id_yield())
print("Nodes: ", len(tree2.nodes()))
print("full yield: ", len(tree2.full_yield()))
raise Exception()
yield tree2
def test_recursive_partitioning_transformation(self):
tree = HybridTree("mytree")
ids = ['a', 'b', 'c', 'd']
for f in ids:
tree.add_node(f, CoNLLToken(f, '_', '_', '_', '_', '_'), True,
True)
if f != 'a':
tree.add_child('a', f)
tree.add_to_root('a')
print(tree)
self.assertEqual([token.form() for token in tree.token_yield()], ids)
self.assertEqual(tree.recursive_partitioning(),
(set([0, 1, 2, 3]), [(set([0]), []), (set([1]), []),
(set([2]), []), (set([3]), [])]))
print(tree.recursive_partitioning())
[fanout_1
] = the_recursive_partitioning_factory().get_partitioning('fanout-1')
print(fanout_1(tree))
def derivation_to_hybrid_tree(der,
poss,
ordered_labels,
construct_token,
disconnected=None):
"""
:param der:
:type der: LCFRSDerivation
:param poss: list of POS-tags
:type poss: list[str]
:param ordered_labels: list of words
:type ordered_labels: list[str]
:param disconnected: list of positions in ordered_labels that are disconnected
:type disconnected: list[object]
:rtype: GeneralHybridTree
Turn a derivation tree into a hybrid tree. Assuming poss and ordered_labels to have equal length.
"""
if not disconnected:
disconnected = []
tree = HybridTree()
j = 1
for i in range(len(ordered_labels)):
token = construct_token(ordered_labels[i], poss[i], True)
if i in disconnected:
tree.add_node("d" + str(i), token, True, False)
else:
tree.add_node("c" + str(j), token, True, True)
j += 1
for id in der.ids():
token = construct_token(der.getRule(id).lhs().nont(), '_', False)
tree.add_node(id, token)
for child in der.child_ids(id):
tree.add_child(id, child)
for position in der.terminal_positions(id):
tree.add_child(id, "c" + str(position))
tree.add_to_root(der.root_id())
tree.reorder()
return tree
def query_result_tree(connection, exp, tree_id):
"""
:param connection:
:param exp:
:param tree_id:
:rtype: str, HybridTree
:return:
"""
cursor = connection.cursor()
result_tree_ids = cursor.execute(
'''SELECT rt_id, status FROM result_trees WHERE exp_id = ? AND t_id = ?''',
(exp, tree_id)).fetchall()
# parse:
if result_tree_ids:
assert (len(result_tree_ids) == 1)
result_tree_id, status = result_tree_ids[0]
if status in ["parse", "fallback"]:
name = cursor.execute('''SELECT name FROM trees WHERE t_id = ?''',
(tree_id, )).fetchall()[0][0]
tree_nodes = cursor.execute((
' SELECT tree_nodes.sent_position, label, pos, result_tree_nodes.head, result_tree_nodes.deprel FROM result_tree_nodes\n'
' JOIN result_trees\n'
' ON result_tree_nodes.rt_id = result_trees.rt_id\n'
' JOIN tree_nodes\n'
' ON result_trees.t_id = tree_nodes.t_id\n'
' AND result_tree_nodes.sent_position = tree_nodes.sent_position\n'
' WHERE result_tree_nodes.rt_id = ?'),
(result_tree_id, ))
tree = HybridTree(name)
for i, label, pos, head, deprel in tree_nodes:
if deprel is None:
deprel = 'UNKNOWN'
token = CoNLLToken(label, '_', pos, pos, '_', deprel)
tree.add_node(str(i), token, True, True)
if head == 0:
tree.add_to_root(str(i))
else:
tree.add_child(str(head), str(i))
assert tree.root is not []
return status, tree
# legacy: no entry found
else:
status = "simple_fallback"
# Create a left branching tree without labels as default strategy
tree_nodes = cursor.execute(
''' SELECT tree_nodes.sent_position, label, pos FROM tree_nodes
WHERE tree_nodes.t_id = ?''', (tree_id, )).fetchall()
left_branch = lambda x: x - 1
right_branch = lambda x: x + 1
strategy = left_branch
length = len(tree_nodes)
tree = HybridTree()
for i, label, pos in tree_nodes:
token = CoNLLToken(label, '_', pos, pos, '_', '_')
tree.add_node(str(i), token, True, True)
parent = strategy(i)
if (parent == 0
and strategy == left_branch) or (parent == length + 1
and strategy == right_branch):
tree.add_to_root(str(i))
else:
tree.add_child(str(parent), str(i))
assert tree.root is not []
return status, tree
def parse_conll_corpus(path, ignore_punctuation, limit=sys.maxsize, start=0):
"""
:param path: path to corpus
:type: str
:param ignore_punctuation: exclude punctuation from tree structure
:type ignore_punctuation: bool
:param limit: stop generation after limit trees
:type: int
:param start: start generation with start'th tree
:type start: int
:return: a series of hybrid trees read from file
:rtype: __generator[HybridTree]
:raise Exception: unexpected input in corpus file
Lazily parses a dependency corpus (in CoNLL format) and generates GeneralHybridTrees.
"""
# print path
with open(path) as file_content:
tree_count = 0
while tree_count < limit:
tree = None
try:
line = next(file_content)
while line.startswith('#'):
line = next(file_content)
except StopIteration:
break
match = CONLL_LINE.match(line)
while match:
if match.group(1) == '1':
tree_count += 1
tree = HybridTree('tree' + str(tree_count))
node_id = match.group(1)
form = match.group(2)
lemma = match.group(3)
cpos = match.group(4)
pos = match.group(5)
feats = match.group(6)
parent = match.group(7)
deprel = match.group(8)
# We ignore information about multiple token's as present in the UD version of Prague Dep. TB
if MULTI_TOKEN.search(node_id):
pass
else:
# If punctuation is to be ignored, we
# remove it from the hybrid tree
# Punctuation according to definition
# cf. http://ilk.uvt.nl/conll/software.html#eval
# if not ignore_punctuation or form.translate(no_translation, string.punctuation):
tree.add_node(node_id, CoNLLToken(form, lemma, cpos, pos, feats, deprel), True, True)
if parent != '0':
tree.add_child(parent, node_id)
# else:
# tree.add_node(node_id, CoNLLToken(form, lemma, pos, fine_grained_pos, feats, deprel), True, False)
# TODO: If punctuation is ignored and the root is punctuation,
# TODO: it is added to the tree anyhow.
if parent == '0':
tree.add_to_root(node_id)
try:
line = next(file_content)
while line.startswith('#'):
line = next(file_content)
match = CONLL_LINE.search(line)
except StopIteration:
line = ''
match = None
# Assume empty line, otherwise raise exception
match = EMPTY_LINE.match(line)
if not match:
raise Exception("Unexpected input in CoNLL corpus file.")
if tree:
# basic sanity checks
if not tree.root:
# FIXME: ignoring punctuation may leads to malformed trees
print("non-rooted")
if ignore_punctuation:
continue
raise Exception
# elif root > 1:
# FIXME: turkish corpus contains trees with more than one root
# FIXME: currently, they are ignored
# continue
elif tree.n_nodes() != len(tree.id_yield()) or len(tree.nodes()) != len(tree.full_yield()):
# FIXME: ignoring punctuation may leads to malformed trees
if ignore_punctuation:
continue
raise Exception(
'{4}: connected nodes: {0}, total nodes: {1}, full yield: {2}, connected yield: {3}'.format(
str(tree.n_nodes()), str(len(tree.nodes())), str(len(tree.full_yield())),
str(len(tree.id_yield())), tree.sent_label()))
if tree_count > start:
yield tree
def parse_with_pgf(grammar, forms, poss, bin):
""""
:type grammar: PGF
:return:
:rtype:
"""
lcfrs = grammar.languages[bin + 'grammargfconcrete']
# sentence = "ADJD ADV _COMMA_ KOUS ADV PIS PROAV VVINF VMFIN _PUNCT_"
sentence = ' '.join(map(escape, poss))
try:
i = lcfrs.parse(sentence, n=1)
p, e = next(i)
except (StopIteration, pgf.ParseError):
return None
# print_ast(gr, e, 0)
s = lcfrs.graphvizParseTree(e)
assert isinstance(s, str)
s_ = s.splitlines()
tree = HybridTree()
# print s
i = 0
for line in s.splitlines():
match = re.search(r'^\s*(n\d+)\[label="([^\s]+)"\]\s*$', line)
if match:
node_id = match.group(1)
label = match.group(2)
order = int(node_id[1:]) >= 100000
if order:
assert escape(poss[i]) == label
tree.add_node(
node_id,
construct_constituent_token(form=forms[i],
pos=poss[i],
terminal=True), True)
i += 1
else:
tree.add_node(
node_id,
construct_constituent_token(form=label,
pos='_',
terminal=False), False)
# print node_id, label
if label == 'VROOT1':
tree.add_to_root(node_id)
continue
match = re.search(r'^ (n\d+) -- (n\d+)\s*$', line)
if match:
parent = match.group(1)
child = match.group(2)
tree.add_child(parent, child)
# print line
# print parent, child
continue
# print tree
assert poss == [token.pos() for token in tree.token_yield()]
# print the_yield
dep_tree = HybridTree()
head_table = defaultdict(lambda: None)
attachment_point = defaultdict(lambda: None)
for i, node in enumerate(tree.id_yield()):
token = tree.node_token(node)
dep_token = construct_conll_token(token.form(), un_escape(token.pos()))
current = tree.parent(node)
current = tree.parent(current)
while current:
current_label = tree.node_token(current).category()
if not re.search(r'\d+X\d+$', current_label):
s = un_escape(current_label)
if s == 'TOP1':
s = 'ROOT1'
dep_token.set_edge_label(s[:-1])
head_table[current] = i + 1
attachment_point[node] = current
break
else:
current = tree.parent(current)
dep_tree.add_node(i + 1, dep_token, order=True)
# print head_table
for node, dep_node in zip(tree.id_yield(), dep_tree.id_yield()):
node = tree.parent(attachment_point[node])
while node:
if head_table[node]:
dep_tree.add_child(head_table[node], dep_node)
break
node = tree.parent(node)
if not node:
dep_tree.add_to_root(dep_node)
# print "dep_tree"
# print dep_tree
# print ' '.join(['(' + token.form() + '/' + token.deprel() + ')' for token in dep_tree.token_yield()])
return dep_tree
class GeneralHybridTreeTestCase(unittest.TestCase):
tree = None
def setUp(self):
self.tree = HybridTree()
self.tree.add_node("v1", construct_conll_token("Piet", "NP"), True)
self.tree.add_node("v21", construct_conll_token("Marie", "N"), True)
self.tree.add_node("v", construct_conll_token("helpen", "VP"), True)
self.tree.add_node("v2", construct_conll_token("lezen", "V"), True)
self.tree.add_child("v", "v2")
self.tree.add_child("v", "v1")
self.tree.add_child("v2", "v21")
self.tree.add_node("v3", construct_conll_token(".", "Punc"), True,
False)
self.tree.add_to_root("v")
def test_children(self):
self.assertListEqual(self.tree.children('v'), ['v2', 'v1'])
self.tree.reorder()
self.assertListEqual(self.tree.children('v'), ['v1', 'v2'])
def test_fringe(self):
self.tree.reorder()
self.assertListEqual(self.tree.fringe('v'), [2, 0, 3, 1])
self.assertListEqual(self.tree.fringe('v2'), [3, 1])
def test_n_spans(self):
self.tree.reorder()
self.assertEqual(self.tree.n_spans('v'), 1)
self.assertEqual(self.tree.n_spans('v2'), 2)
def test_n_gaps(self):
self.tree.reorder()
self.assertEqual(self.tree.n_gaps(), 1)
def test_node_ids(self):
self.tree.reorder()
self.assertListEqual(sorted(self.tree.nodes()),
sorted(['v', 'v1', 'v2', 'v21', 'v3']))
def test_complete(self):
self.tree.reorder()
self.assertEqual(self.tree.complete(), True)
def test_unlabelled_structure(self):
self.tree.reorder()
self.assertTupleEqual(self.tree.unlabelled_structure(),
({0, 1, 2, 3}, [({0}, []),
({1, 3}, [({1}, [])])]))
def test_max_n_spans(self):
self.tree.reorder()
self.assertEqual(self.tree.max_n_spans(), 2)
def test_labelled_yield(self):
self.tree.reorder()
self.assertListEqual(
[token.form() for token in self.tree.token_yield()],
"Piet Marie helpen lezen".split(' '))
def test_full_labelled_yield(self):
self.tree.reorder()
self.assertListEqual(
[token.form() for token in self.tree.full_token_yield()],
"Piet Marie helpen lezen .".split(' '))
def test_full_yield(self):
self.tree.reorder()
self.assertListEqual(self.tree.full_yield(),
'v1 v21 v v2 v3'.split(' '))
# def test_labelled_spans(self):
# self.tree.reorder()
# self.assertListEqual(self.tree.labelled_spans(), [])
def test_pos_yield(self):
self.tree.reorder()
self.assertListEqual(
[token.pos() for token in self.tree.token_yield()],
"NP N VP V".split(' '))
def test_recursive_partitioning(self):
self.tree.reorder()
self.assertEqual(self.tree.recursive_partitioning(),
({0, 1, 2, 3}, [({0}, []),
({1, 3}, [({1}, []), ({3}, [])]),
({2}, [])]))