def acl_mod_verb(dep_graph: DependencyGraph, oia_graph: OIAGraph,
context: UD2OIAContext):
"""
this is called after adnominal_clause_mark, which means there is no mark
:param dep_graph:
:param oia_graph:
:return:
"""
pattern = DependencyGraph()
noun_node = pattern.create_node(UPOS="NOUN|PRON|PROPN|ADJ|ADV|NUM")
# ADJ is for the cases that "many/some" are abbrv of many X/some X, representing NOUN
# ADV is for the case of "here" for "i am here thinking xxx"
verb_node = pattern.create_node(UPOS="VERB|AUX")
# aux is for can, have which ommits the true verb
pattern.add_nodes([noun_node, verb_node])
pattern.add_dependency(noun_node, verb_node, r'acl')
for match in dep_graph.match(pattern):
dep_noun_node = match[noun_node]
dep_verb_node = match[verb_node]
if context.is_processed(dep_noun_node, dep_verb_node):
continue
if oia_graph.has_relation(dep_noun_node,
dep_verb_node,
direct_link=False):
continue
oia_verb_node = oia_graph.add_words(dep_verb_node.position)
oia_noun_node = oia_graph.add_words(dep_noun_node.position)
dep = dep_graph.get_dependency(dep_noun_node, dep_verb_node)
labels = [x for x in dep.rels if x.startswith("acl:")]
pred = None
if labels:
assert len(labels) == 1
label = labels[0]
pred = label.split(":")[1]
if pred == "relcl":
pred = None
# if pred:
# # there is no mark, but we add it because it may be because of not being shared in conjunction
#
# oia_pred_node = oia_graph.add_aux(pred)
# oia_graph.add_argument(oia_pred_node, oia_noun_node, 1, mod=True)
# oia_graph.add_argument(oia_pred_node, oia_verb_node, 2)
# else:
oia_graph.add_mod(oia_verb_node, oia_noun_node)
def amod_obl(dep_graph: DependencyGraph):
"""
##### include: more than, successful by
:param dep_graph:
:param oia_graph:
:return:
"""
pattern = DependencyGraph()
noun_node = DependencyGraphNode(UPOS=r"NOUN|PRON")
adj_node = DependencyGraphNode(UPOS="ADJ")
adp_node = DependencyGraphNode(UPOS="ADP")
obl_node = DependencyGraphNode()
pattern.add_nodes([noun_node, adj_node, adp_node, obl_node])
pattern.add_dependency(noun_node, adj_node, r'amod')
pattern.add_dependency(adj_node, obl_node, r'obl:\w+')
pattern.add_dependency(obl_node, adp_node, r'case')
more_than_pred = []
for match in dep_graph.match(pattern):
dep_noun_node = match[noun_node]
dep_adj_node = match[adj_node]
dep_obl_node = match[obl_node]
dep_adp_node = match[adp_node]
obl_nodes = list(
dep_graph.children(dep_adj_node, filter=lambda n, l: "obl" in l))
if len(obl_nodes) > 1:
# similar in form to the one
continue
if dep_adp_node.FORM not in dep_graph.get_dependency(
dep_adj_node, dep_obl_node).values():
continue
if dep_noun_node.LOC < dep_adj_node.LOC < dep_adp_node.LOC < dep_obl_node.LOC:
more_than_pred.append(
(dep_noun_node, dep_adj_node, dep_obl_node, dep_adp_node))
for dep_noun_node, dep_adj_node, dep_obl_node, dep_adp_node in more_than_pred:
nodes = [dep_adj_node, dep_adp_node]
more_than_pred = merge_dep_nodes(nodes,
UPOS="ADP",
LOC=dep_adp_node.LOC)
dep_graph.remove_dependency(dep_noun_node, dep_adj_node)
dep_graph.remove_dependency(dep_adj_node, dep_obl_node)
dep_graph.replace_nodes([dep_adj_node, dep_adp_node], more_than_pred)
dep_graph.add_dependency(dep_noun_node, dep_obl_node,
"nmod:" + more_than_pred.FORM)
def and_or(dep_graph: DependencyGraph):
"""
:param dep_graph:
:param oia_graph:
:return:
"""
pattern = DependencyGraph()
parent_node = pattern.create_node()
some_node = pattern.create_node()
and_node = pattern.create_node(LEMMA=r"\band\b")
or_node = pattern.create_node(LEMMA=r"\bor\b")
pattern.add_dependency(parent_node, some_node, r'\bconj:\w*')
pattern.add_dependency(some_node, and_node, r'\bcc\b')
pattern.add_dependency(some_node, or_node, r'\bcc\b')
pattern.add_dependency(and_node, or_node, r'\bconj')
for match in list(dep_graph.match(pattern)):
dep_parent_node = match[parent_node]
dep_some_node = match[some_node]
dep_and_node = match[and_node]
dep_or_node = match[or_node]
rel = dep_graph.get_dependency(dep_parent_node, dep_some_node)
if not rel.startswith("conj:and") and not rel.startswith("conj:or"):
continue
and_or_nodes = [n for n in dep_graph.nodes() if dep_and_node.LOC < n.LOC < dep_or_node.LOC]
if any([node.UPOS in {"VERB", "NOUN", "ADJ", "ADP", "ADV"} for node in and_or_nodes]):
continue
and_or_nodes.append(dep_and_node)
and_or_nodes.append(dep_or_node)
and_or_nodes.sort(key=lambda n: n.LOC)
if not all([dep_graph.get_node(x.ID) for x in and_or_nodes]):
continue
new_and_or_node = merge_dep_nodes(and_or_nodes,
UPOS=dep_and_node.UPOS,
LOC=dep_and_node.LOC,
FEATS=dep_and_node.FEATS
)
dep_graph.replace_nodes(and_or_nodes, new_and_or_node)
dep_graph.set_dependency(dep_parent_node, dep_some_node, "conj:" + new_and_or_node.FORM)
def multi_word_sconj(dep_graph: DependencyGraph):
"""
:param dep_graph:
:param oia_graph:
:return:
"""
pattern = DependencyGraph()
verb_node = pattern.create_node(UPOS="VERB")
verb2_node = pattern.create_node(UPOS="VERB")
mark_node = pattern.create_node(UPOS="SCONJ")
pattern.add_dependency(verb_node, verb2_node, r'advcl:\w*')
pattern.add_dependency(verb2_node, mark_node, r'mark')
mark_phrases = []
for match in dep_graph.match(pattern):
dep_verb_node = match[verb_node]
dep_verb2_node = match[verb2_node]
dep_mark_node = match[mark_node]
if dep_mark_node.LEMMA not in dep_graph.get_dependency(dep_verb_node, dep_verb2_node).values():
continue
new_marks = list(dep_graph.offsprings(dep_mark_node))
if len(new_marks) == 1:
continue
new_marks.sort(key=lambda n: n.LOC)
mark_phrases.append((dep_verb_node, dep_verb2_node, dep_mark_node, new_marks))
for (dep_verb_node, dep_verb2_node, dep_mark_node, new_marks) in mark_phrases:
if not all([dep_graph.get_node(x.ID) for x in new_marks]):
continue
dep_graph.remove_dependency(dep_verb2_node, dep_mark_node)
dep_graph.remove_dependency(dep_verb_node, dep_verb2_node)
new_mark_node = merge_dep_nodes(new_marks,
UPOS=dep_mark_node.UPOS,
LOC=dep_mark_node.LOC
)
dep_graph.replace_nodes(new_marks, new_mark_node)
dep_graph.add_dependency(dep_verb_node, dep_verb2_node, "advcl:" + new_mark_node.LEMMA)
dep_graph.add_dependency(dep_verb2_node, new_mark_node, "mark")
def acl_loop(dep_graph: DependencyGraph):
"""
:param dep_graph:
:param oia_graph:
:return:
"""
for n1, n2, deps in dep_graph.dependencies():
if "acl:relcl" in deps:
back_deps = dep_graph.get_dependency(n2, n1)
if any(x in back_deps
for x in {"obl", "nsubj", "obj", "mark", "advmod"}):
dep_graph.remove_dependency(n2, n1)
def nmod_with_case(dep_graph: DependencyGraph, oia_graph: OIAGraph,
context: UD2OIAContext):
"""
#################### nmod:x ########################
##### the office of the chair #####
##### Istanbul in Turkey #####
:param sentence:
:return:
"""
pattern = DependencyGraph()
parent_node = DependencyGraphNode()
child_node = DependencyGraphNode()
case_node = DependencyGraphNode()
pattern.add_nodes([parent_node, child_node, case_node])
pattern.add_dependency(parent_node, child_node, r'\w*nmod\w*')
pattern.add_dependency(child_node, case_node, r'\w*case\w*')
for match in dep_graph.match(pattern):
dep_parent_node = match[parent_node]
dep_child_node = match[child_node]
dep_case_node = match[case_node]
rel = dep_graph.get_dependency(dep_parent_node, dep_child_node)
# vs, lemma = versus
# according, lemma = accord,
# but rel always select the shorter one
if oia_graph.has_relation(dep_parent_node, dep_child_node):
continue
if rel != "nmod:" + dep_case_node.LEMMA and rel != 'nmod:' + dep_case_node.FORM:
pred_node = oia_graph.add_words(dep_case_node.position)
else:
pred_node = oia_graph.add_words(dep_case_node.position)
arg1_node = oia_graph.add_words(dep_parent_node.position)
arg2_node = oia_graph.add_words(dep_child_node.position)
oia_graph.add_argument(pred_node, arg1_node, 1, mod=True)
oia_graph.add_argument(pred_node, arg2_node, 2)
def get_relation_to_conj(dep_graph: DependencyGraph, root, root_parents,
parallel_components):
"""
:param dep_graph:
:param parallel_components:
:return:
"""
relation_to_conj = dict()
for parent in root_parents:
prefixs = []
marks = []
shared_prefix = True
for child in parallel_components:
rels = dep_graph.get_dependency(parent, child)
if rels:
rel = list(rels.rels)[0]
if child != root and rel.startswith("conj"):
continue
if ":" in rel:
prefix, mark = rel.split(":")
if mark in {"relcl", "xsubj", "pass", "poss", "tmod"}:
prefix = rel
mark = None
else:
prefix = rel
mark = None
prefixs.append(prefix)
marks.append(mark)
else:
shared_prefix = False
marks.append(None)
# ic(str(parent))
# ic(list(map(str, parallel_components)))
assert (len(set(prefixs))) == 1
prefix = prefixs[0]
if all([m is None for m in marks]):
marks = None
if any(x in prefix for x in {"subj", "obj", "ccomp", "xcomp"}):
marks = None
relation_to_conj[parent.ID] = (prefix, shared_prefix, marks)
return relation_to_conj
def oblique_without_prep(dep_graph: DependencyGraph, oia_graph: OIAGraph,
context: UD2OIAContext):
"""
:param dep_graph:
:param oia_graph:
:return:
"""
# cut X by a knife
pattern = DependencyGraph()
verb_node = DependencyGraphNode(UPOS="VERB|NOUN|ADJ|PROPN|PRON")
oblique_node = DependencyGraphNode()
pattern.add_node(verb_node)
pattern.add_node(oblique_node)
pattern.add_dependency(verb_node, oblique_node, r'obl:tmod|obl:npmod|obl')
for match in dep_graph.match(pattern):
dep_verb_node = match[verb_node]
dep_oblique_node = match[oblique_node]
if oia_graph.has_relation(dep_verb_node,
dep_oblique_node,
direct_link=False):
continue
oblique_edge = dep_graph.get_dependency(dep_verb_node,
dep_oblique_node)
oblique_types = oblique_edge.values()
if "tmod" in oblique_types:
oia_pred_node = oia_graph.add_aux("TIME_IN")
arg1_node = oia_graph.add_words(dep_verb_node.position)
arg2_node = oia_graph.add_words(dep_oblique_node.position)
oia_graph.add_argument(oia_pred_node, arg1_node, 1, mod=True)
oia_graph.add_argument(oia_pred_node, arg2_node, 2)
else: # "npmod" in oblique_types and others
oia_verb_node = oia_graph.add_words(dep_verb_node.position)
obl_node = oia_graph.add_words(dep_oblique_node.position)
oia_graph.add_mod(obl_node, oia_verb_node)
def nmod_without_case(dep_graph: DependencyGraph, oia_graph: OIAGraph,
context: UD2OIAContext):
"""
#################### nmod:x ########################
:param sentence:
:return:
"""
pattern = DependencyGraph()
center_node = pattern.create_node()
modifier_node = pattern.create_node()
pattern.add_dependency(center_node, modifier_node, r'\w*nmod\w*')
for match in dep_graph.match(pattern):
dep_center_node = match[center_node]
dep_modifier_node = match[modifier_node]
rels = dep_graph.get_dependency(dep_center_node, dep_modifier_node)
if "nmod:poss" in rels and dep_center_node in set(
dep_graph.offsprings(dep_modifier_node)):
# whose in there
continue
if oia_graph.has_relation(dep_center_node,
dep_modifier_node,
direct_link=False):
continue
oia_center_node = oia_graph.add_words(dep_center_node.position)
oia_modifier_node = oia_graph.add_words(dep_modifier_node.position)
oia_graph.add_mod(oia_modifier_node, oia_center_node)
def two_node_with_case(dep_graph: DependencyGraph, oia_graph: OIAGraph,
context: UD2OIAContext):
"""
:param dep_graph:
:param oia_graph:
:return:
"""
regular_nodes = [
n for n in dep_graph.nodes() if n.UPOS not in {"ROOT", "PUNCT"}
]
#logger.debug("regular nodes")
#for node in regular_nodes:
# logger.debug(str(node))
if len(regular_nodes) == 2:
regular_nodes.sort(key=lambda x: x.LOC)
case_node, noun_node = regular_nodes
if dep_graph.get_dependency(noun_node, case_node) == "case":
oia_case_node = oia_graph.add_words(case_node.position)
oia_noun_node = oia_graph.add_words(noun_node.position)
oia_graph.add_argument(oia_case_node, oia_noun_node, 2)
def acl_verb_obl_case(dep_graph: DependencyGraph):
"""
something extracted by
:param dep_graph:
:param oia_graph:
:return:
"""
pattern = DependencyGraph()
subj_node = pattern.create_node()
verb_node = pattern.create_node(UPOS="VERB")
obj_node = pattern.create_node()
case_node = pattern.create_node()
pattern.add_dependency(subj_node, verb_node, r'acl')
pattern.add_dependency(verb_node, obj_node, r'obl:\w*')
pattern.add_dependency(obj_node, case_node, r'case')
phrases = []
for match in dep_graph.match(pattern):
dep_subj_node = match[subj_node]
dep_verb_node = match[verb_node]
dep_obj_node = match[obj_node]
dep_case_node = match[case_node]
obl_nodes = [
n for n, l in dep_graph.children(
dep_verb_node, filter=lambda n, l: l.startswith("obl"))
]
if len(obl_nodes) > 1:
continue
existing_obj_nodes = [
n for n, l in dep_graph.children(
dep_verb_node, filter=lambda n, l: "obj" in l or "comp" in l)
]
if existing_obj_nodes:
continue
obl_rel = dep_graph.get_dependency(dep_verb_node, dep_obj_node)
if dep_case_node.FORM not in obl_rel.values():
continue
# there are may be other cases, join them all
dep_case_nodes = [
n for n, l in
dep_graph.children(dep_obj_node,
filter=lambda n, l: l.startswith("case") and
dep_verb_node.LOC < n.LOC < dep_obj_node.LOC)
]
subjs = list(
dep_graph.children(dep_verb_node, filter=lambda n, l: "subj" in l))
if len(subjs) > 1:
continue
phrases.append(
(dep_subj_node, dep_verb_node, dep_obj_node, dep_case_nodes))
for dep_subj_node, dep_verb_node, dep_obj_node, dep_case_nodes in phrases:
new_verb_phrase = [dep_verb_node] + dep_case_nodes
logging.debug("acl_verb_obl_case: we are merging nodes")
logging.debug("\n".join(str(node) for node in new_verb_phrase))
new_verb_node = merge_dep_nodes(new_verb_phrase,
UPOS=dep_verb_node.UPOS,
LOC=dep_verb_node.LOC,
FEATS=dep_verb_node.FEATS)
logging.debug("acl_verb_obl_case: we obtain a new node")
logging.debug(str(new_verb_node))
dep_graph.remove_dependency(dep_verb_node, dep_obj_node)
for node in dep_case_nodes:
dep_graph.remove_dependency(dep_obj_node, node)
dep_graph.replace_nodes(new_verb_phrase, new_verb_node)
dep_graph.add_dependency(new_verb_node, dep_obj_node, "obj")
def object_relative_clause(dep_graph: DependencyGraph, oia_graph: OIAGraph,
context: UD2OIAContext):
"""
##### Object-extracted/referred relative clause #####
##### the person that Andy knows #####
:param sentence:
:return:
"""
pattern = DependencyGraph()
verb_node = DependencyGraphNode()
entity_node = DependencyGraphNode()
subj_node = DependencyGraphNode()
pattern.add_nodes([verb_node, entity_node, subj_node])
pattern.add_dependency(verb_node, subj_node, r'\w*subj\w*')
pattern.add_dependency(entity_node, verb_node, r'\w*acl:relcl\w*')
for match in dep_graph.match(pattern):
dep_entity_node = match[entity_node]
dep_subj_node = match[subj_node]
dep_verb_node = match[verb_node]
if dep_subj_node.LEMMA in {"what", "who", "which", "that"}:
continue
logger.debug("we found a objective relative clause")
logger.debug("entity: {0}".format(dep_entity_node))
logger.debug("subject: {0}".format(dep_subj_node))
logger.debug("verb: {0}".format(dep_verb_node))
if context.is_processed(dep_entity_node, dep_verb_node):
logger.debug("processed")
continue
context.processed(dep_verb_node, dep_subj_node)
context.processed(dep_entity_node, dep_verb_node)
oia_entity_node = oia_graph.add_words(dep_entity_node.position)
oia_verb_node = oia_graph.add_words(dep_verb_node.position)
oia_subj_node = oia_graph.add_words(dep_subj_node.position)
if oia_graph.has_relation(oia_entity_node, oia_verb_node):
logger.debug("has relation between entity and verb")
continue
oia_graph.add_argument(oia_verb_node, oia_subj_node, 1)
def __valid_ref(n, l):
return l == "ref" and dep_entity_node.LOC < n.LOC < dep_verb_node.LOC
ref_nodes = list(n for n, l in dep_graph.children(dep_entity_node,
filter=__valid_ref))
ref_nodes.sort(key=lambda x: x.LOC)
if ref_nodes:
ref_node = ref_nodes[-1]
oia_ref_node = oia_graph.add_words(ref_node.position)
oia_graph.add_ref(oia_entity_node, oia_ref_node)
logger.debug("we are coping with ref between:")
logger.debug(dep_verb_node)
logger.debug(ref_node)
ref_relation = dep_graph.get_dependency(dep_verb_node, ref_node)
case_nodes = list(n for n, l in dep_graph.children(
ref_node, filter=lambda n, l: "case" in l))
case_nodes.sort(key=lambda x: x.LOC)
if ref_relation:
if case_nodes:
# with which xxxx, the with will become the root pred
case_node = case_nodes[-1]
oia_case_node = oia_graph.add_words(case_node.position)
oia_graph.add_argument(oia_case_node,
oia_verb_node,
1,
mod=True)
oia_graph.add_argument(oia_case_node, oia_ref_node, 2)
oia_graph.add_mod(oia_verb_node, oia_entity_node)
else:
if "obj" in ref_relation:
oia_graph.add_argument(oia_verb_node, oia_ref_node, 2)
elif ref_relation == "advmod":
oia_graph.add_mod(oia_ref_node, oia_verb_node)
else:
raise Exception(
"unknown relation: {}".format(ref_relation))
# oia_graph.add_argument(oia_verb_node, oia_entity_node, 2, mod=True)
oia_graph.add_argument(oia_verb_node, oia_subj_node, 1)
oia_graph.add_argument(oia_verb_node, oia_entity_node, 2, mod=True)
rels = dep_graph.get_dependency(dep_entity_node, dep_verb_node)
#if rels.endswith("obj"):
for node, l in dep_graph.children(dep_verb_node):
if l == "ccomp":
oia_ccomp_node = oia_graph.add_words(node.position)
oia_graph.add_argument(oia_verb_node, oia_ccomp_node, 3)
def verb_phrase(dep_graph: DependencyGraph):
"""
##### Merging aux and cop with their head VERB #####
Cases:
:param sentence:
:return:
"""
verb_phrases = []
for node in dep_graph.nodes(filter=lambda x: x.UPOS in {"VERB", "AUX"}):
if node.UPOS == "AUX":
parent = [
n for n, l in dep_graph.parents(node,
filter=lambda n, l: l == "aux")
]
if len(parent) > 0:
continue
# if "VerbForm" in node.FEATS and "Ger" in node.FEATS["VerbForm"]:
# continue
if "Tense" in node.FEATS and "Past" in node.FEATS["Tense"]:
# if the verb is before the noun, it will be processed by noun_phrase and taken as a part of the noun
parent = [
n for n, l in dep_graph.parents(
node, filter=lambda n, l: l == "amod" and node.LOC < n.LOC)
]
if len(parent) > 0:
continue
# logger.debug("We are checking node {0}".format(node))
root = node
verbs = [root]
for n, l in dep_graph.children(root):
if dep_graph.get_dependency(n, root):
continue
if n.LEMMA in {"so", "also", "why"}:
continue
if "advmod" in l:
offsprings = list(dep_graph.offsprings(n))
if any(x.UPOS in {"VERB", "NOUN", "AUX", "PRON"}
for x in offsprings):
continue
verbs.extend(offsprings)
elif "compound" in l:
verbs.append(n)
verbs = [
x for x in verbs if x.LOC <= root.LOC
or "compound" in dep_graph.get_dependency(root, x)
]
# logger.debug("Verb: before continuous component ")
# logger.debug("\n".join(str(verb) for verb in verbs))
verbs = continuous_component(verbs, root)
# add aux
verbs.extend(n for n, l in dep_graph.children(root) if "aux" in l)
# logger.debug("Verb: after continuous component ")
# for verb in verbs:
# logger.debug(verb)
verbs.sort(key=lambda x: x.LOC)
last_loc = verbs[-1].LOC
# next_node = dep_graph.get_node_by_loc(last_loc + 1)
# if next_node and next_node.LEMMA == "not":
# verbs.append(next_node)
if len(verbs) > 1:
verb_phrases.append((verbs, root))
for verbs, root in verb_phrases:
verb_node = merge_dep_nodes(verbs,
UPOS="VERB",
LOC=root.LOC,
FEATS=root.FEATS)
dep_graph.replace_nodes(verbs, verb_node)
def multi_words_case(dep_graph: DependencyGraph):
"""
:TODO add example case
:param dep_graph:
:param oia_graph:
:return:
"""
pattern = DependencyGraph()
noun_node = DependencyGraphNode()
x_node = DependencyGraphNode()
case_node = DependencyGraphNode()
pattern.add_node(noun_node)
pattern.add_node(x_node)
pattern.add_node(case_node)
pattern.add_dependency(noun_node, x_node, r'\w*:\w*')
pattern.add_dependency(x_node, case_node, r'\bcase\b')
for match in list(dep_graph.match(pattern)):
multiword_cases = []
dep_noun_node = match[noun_node]
dep_x_node = match[x_node]
dep_case_node = match[case_node]
if not dep_graph.has_node(dep_case_node):
continue
direct_case_nodes = [n for n, l in dep_graph.children(dep_x_node, filter=lambda n, l: "case" == l)]
all_case_nodes = set()
for node in direct_case_nodes:
all_case_nodes.update(dep_graph.offsprings(node))
if len(all_case_nodes) == 1:
continue
all_case_nodes = sorted(list(all_case_nodes), key=lambda n: n.LOC)
logger.debug("multi case discovered")
for node in all_case_nodes:
logger.debug(str(node))
# if len(case_nodes) > 2:
# raise Exception("multi_words_case: Unexpected Situation: nodes with more than two cases")
x_rel = dep_graph.get_dependency(dep_noun_node, dep_x_node)
for rel in x_rel:
if ":" in rel:
# print('-----------------rel: ',rel)
rel_str, case_str = rel.split(":")
# some times, the rel only contains one word
# Example :
# that OBSF values within the extended trial balance may be misstated due to data issues ( above and beyond existing conversations with AA on model simplifications)
if case_str in "_".join([x.LEMMA for x in all_case_nodes]):
multiword_cases.append((dep_noun_node, dep_x_node, dep_case_node, all_case_nodes, rel_str))
for dep_noun_node, dep_x_node, dep_case_node, case_nodes, rel_str in multiword_cases:
logger.debug("we are merging:")
for node in case_nodes:
logger.debug(str(node))
if not all([dep_graph.has_node(x) for x in case_nodes]):
continue
new_case_node = merge_dep_nodes(case_nodes,
UPOS=dep_case_node.UPOS,
LOC=dep_case_node.LOC
)
dep_graph.replace_nodes(case_nodes, new_case_node)
dep_graph.remove_dependency(dep_noun_node, dep_x_node)
dep_graph.add_dependency(dep_noun_node, dep_x_node,
rel_str + ":" + " ".join([x.LEMMA for x in case_nodes]))
def process_conjunction(dep_graph: DependencyGraph, root: DependencyGraphNode):
"""
:param dep_graph:
:param root:
:return:
"""
conj_childs = [
child for child, rels in dep_graph.children(
root, filter=lambda n, l: l.startswith("conj"))
]
assert conj_childs
parallel_components = [root]
for child in conj_childs:
is_nest = any(
grand_rels.startswith("conj")
for grand_sun, grand_rels in dep_graph.children(child))
if is_nest:
logger.debug("nested conj is found ")
logger.debug(str(child))
conj_node, parallel_nodes = process_conjunction(dep_graph, child)
logger.debug("conj_node is created ")
logger.debug(str(conj_node))
for node in parallel_nodes:
logger.debug("Containing nodes ")
logger.debug(str(node))
rels = list(dep_graph.get_dependency(root, node))
for rel in rels:
if rel.startswith("conj"):
logger.debug("remove dependency {0}".format(
(root.ID, node.ID, rel)))
dep_graph.remove_dependency(root, node, rel)
dep_graph.add_dependency(root, conj_node, rel)
child = conj_node
parallel_components.append(child)
parallel_components.sort(key=lambda x: x.LOC)
# if all(n.UPOS in NOUN_UPOS for n in parallel_components):
#
# logger.debug("Processing all noun conjunction")
#
# is_pure_noun = True
#
# merging_noun_nodes = []
# min_loc = 10000
# max_loc = -1
# for child in parallel_components:
# if isinstance(child, DependencyGraphNode):
# min_loc = min(min_loc, child.LOC)
# max_loc = max(min_loc, child.LOC)
# elif isinstance(child, DependencyGraphSuperNode):
# min_loc = min(min_loc, min([x.LOC for x in child.nodes]))
# max_loc = max(max_loc, max([x.LOC for x in child.nodes]))
# merging_noun_nodes.extend(dep_graph.offsprings(child))
#
# logger.debug("Checking acl for {0}".format(child))
# for n, l in dep_graph.children(child):
# logger.debug(n)
# logger.debug("label {0}".format(l))
# if "acl" in l:
# is_pure_noun = False
# break
#
# if is_pure_noun:
# merging_noun_nodes = [n for n in merging_noun_nodes if min_loc <= n.LOC <= max_loc]
# is_pure_noun = not any(n.UPOS in {"ADP", "VERB", "SCONJ", "AUX"} for n in merging_noun_nodes)
#
# if is_pure_noun:
# # merged_noun_nodes.sort(key=lambda x: x.LOC)
# for node in merging_noun_nodes:
# logger.debug("merging {0}".format(node))
#
# new_noun = merge_dep_nodes(merging_noun_nodes, UPOS=root.UPOS, LOC=root.LOC)
# dep_graph.replace_nodes(merging_noun_nodes, new_noun)
#
# return new_noun, []
root_parents = list(set(parent
for parent, rels in dep_graph.parents(root)))
root_parents.sort(key=lambda x: x.LOC)
# ic(list(map(str, root_parents)))
conj_node, with_arg_palceholder = build_conjunction_node(
dep_graph, root, root_parents, parallel_components)
relation_to_conj = get_relation_to_conj(dep_graph, root, root_parents,
parallel_components)
case_marks = dict()
for index, node in enumerate(parallel_components):
case_marks[node.ID] = [(n, l) for n, l in dep_graph.children(node)
if ("case" in l or "mark" in l or "cc" in l)]
for key, values in case_marks.items():
for v in values:
logger.debug("case_marker = {} {} {}".format(
key, v[0].ID, v[1].rels))
logger.debug("relation_to_conj = {}".format(relation_to_conj))
for parent in root_parents:
# ic(parent)
prefix, shared_prefix, required_mark = relation_to_conj[parent.ID]
if any(x in prefix for x in {"subj", "obj", "ccomp", "xcomp"}) \
or not required_mark or len(set(required_mark)) == 1:
for node in parallel_components:
dep_graph.remove_dependency(parent, node)
relation = prefix
if required_mark and len(set(required_mark)) == 1:
## with same mark
mark_lemma = list(set(required_mark))[0]
relation += ":" + mark_lemma
mark_node = find_mark(case_marks, parallel_components,
mark_lemma)
if mark_node:
mark_node, mark_rel = mark_node
dep_graph.remove_node(mark_node)
dep_graph.add_node(mark_node) # clear the dependency
dep_graph.add_dependency(conj_node, mark_node, mark_rel)
else:
logger.error("cannot find the mark node")
dep_graph.add_dependency(parent, conj_node, relation)
else:
complete_missing_case_mark(dep_graph, root, root_parents,
parallel_components, relation_to_conj,
case_marks)
if not required_mark:
required_mark = [None] * len(parallel_components)
for index, (node, mark) in enumerate(
zip(parallel_components, required_mark)):
if mark:
rel = prefix + ":" + mark
else:
rel = prefix
# if rel.startswith("conj"):
# continue
logger.debug("add dependency {0}".format(
(parent.ID, node.ID, rel)))
dep_graph.add_dependency(parent, node, rel)
for idx, node in enumerate(parallel_components):
if node != root:
rels = dep_graph.get_dependency(root, node)
for rel in rels:
if rel.startswith("conj"):
dep_graph.remove_dependency(root, node)
if with_arg_palceholder:
index = idx + 1
else:
# a, but b, b should be the arg1 and a be the arg2
index = len(parallel_components) - idx
dep_graph.add_dependency(conj_node, node,
"arg_conj:{0}".format(index))
return conj_node, parallel_components
