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 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 build_score_validator(baseline_grammar, grammarInfo, nont_map,
storageManager, term_labelling, parser,
corpus_validation, validationMethod):
validator = PyCandidateScoreValidator(grammarInfo, storageManager,
validationMethod)
# parser = GFParser(baseline_grammar)
tree_count = 0
der_count = 0
for gold_tree in corpus_validation.get_trees():
tree_count += 1
parser.set_input(
term_labelling.prepare_parser_input(gold_tree.token_yield()))
parser.parse()
derivations = map(lambda x: x[1], parser.k_best_derivation_trees())
manager = PyDerivationManager(baseline_grammar, nont_map)
manager.convert_derivations_to_hypergraphs(derivations)
scores = []
gold_labels = {}
gold_heads = {}
for position, id in enumerate(gold_tree.id_yield()):
parent_id = gold_tree.parent(id)
gold_labels[position] = gold_tree.node_token(id).deprel()
if parent_id is None:
assert id in gold_tree.root
gold_heads[position] = 0
else:
gold_heads[position] = gold_tree.id_yield().index(
parent_id) + 1
derivations = parser.k_best_derivation_trees()
for _, der in derivations:
der_count += 1
h_tree = HybridTree()
cleaned_tokens = copy.deepcopy(gold_tree.full_token_yield())
dcp = DCP_evaluator(der).getEvaluation()
dcp_to_hybridtree(h_tree, dcp, cleaned_tokens, False,
construct_conll_token)
las, uas, lac = 0, 0, 0
for position, id in enumerate(h_tree.id_yield()):
parent_id = h_tree.parent(id)
if parent_id is None:
assert id in h_tree.root
head = 0
else:
head = h_tree.id_yield().index(parent_id) + 1
label = h_tree.node_token(id).deprel()
if gold_heads[position] == head:
uas += 1
if gold_labels[position] == label:
lac += 1
if gold_heads[position] == head and gold_labels[
position] == label:
las += 1
if validationMethod == "LAS":
scores.append(las)
elif validationMethod == "UAS":
scores.append(uas)
elif validationMethod == "LAC":
scores.append(lac)
max_score = len(gold_tree.id_yield())
validator.add_scored_candidates(manager, scores, max_score)
print(tree_count, max_score, scores)
parser.clear()
print("trees used for validation ", tree_count, "with",
der_count * 1.0 / tree_count, "derivations on average")
return validator
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