def get_adj_verb_phrase(dep_graph):
"""
:param dep_graph:
:param oia_graph:
:return:
"""
pattern = DependencyGraph()
adj_node = pattern.create_node(UPOS="ADJ")
get_node = pattern.create_node(LEMMA="get", UPOS="VERB")
pattern.add_dependency(adj_node, get_node, r'aux')
verb_phrases = []
for match in dep_graph.match(pattern):
dep_adj_node = match[adj_node]
dep_get_node = match[get_node]
if isinstance(dep_adj_node,
DependencyGraphSuperNode) and dep_adj_node.is_conj:
continue
pred = [dep_get_node, dep_adj_node]
head = dep_adj_node
verb_phrases.append((pred, head))
for verbs, root in verb_phrases:
verb_node = merge_dep_nodes(verbs, UPOS="VERB", LOC=root.LOC)
dep_graph.replace_nodes(verbs, verb_node)
def adj_modifier(dep_graph: DependencyGraph, oia_graph: OIAGraph,
context: UD2OIAContext):
"""
adj previous to noun is coped with by noun phrase
this process the case that adj is behind the noun
#################### a pretty little boy ########################
:param dep_graph:
:param oia_graph:
:return:
"""
pattern = DependencyGraph()
noun_node = pattern.create_node() # UPOS="NOUN|PRON|PROPN")
adj_node = pattern.create_node() # UPOS="ADJ|NOUN")
pattern.add_dependency(noun_node, adj_node, r'amod')
for match in dep_graph.match(pattern):
dep_noun_node = match[noun_node]
dep_adj_node = match[adj_node]
oia_noun_node = oia_graph.add_words(dep_noun_node.position)
oia_adj_node = oia_graph.add_words(dep_adj_node.position)
logger.debug("adj_modifier: ")
logger.debug(dep_noun_node.position)
logger.debug(oia_noun_node)
logger.debug(dep_adj_node.position)
logger.debug(oia_adj_node)
oia_graph.add_mod(oia_adj_node, oia_noun_node)
def reverse_passive_verb(dep_graph: DependencyGraph):
"""
I'd forgotten how blown away I was by some of the songs the first time I saw it in NY.
:param dep_graph:
:return:
"""
pattern = DependencyGraph()
subj_node = pattern.create_node()
verb_node = pattern.create_node(UPOS="VERB", FEATS={"Tense": "Past"})
be_node = pattern.create_node(LEMMA=r"\bbe\b")
pattern.add_dependency(verb_node, subj_node, r"\w*subj")
pattern.add_dependency(verb_node, be_node, "cop")
for match in list(dep_graph.match(pattern)):
dep_subj_node = match[subj_node]
dep_verb_node = match[verb_node]
dep_be_node = match[be_node]
if not (dep_verb_node.LOC < dep_subj_node.LOC < dep_be_node.LOC):
continue
new_verb_phrase = [dep_be_node, dep_verb_node]
dep_new_verb = merge_dep_nodes(new_verb_phrase,
UPOS="VERB",
LOC=dep_be_node.LOC)
dep_graph.replace_nodes(new_verb_phrase, dep_new_verb)
def obl_modifier(dep_graph: DependencyGraph, oia_graph: OIAGraph,
context: UD2OIAContext):
"""
the adv before the verb should be processed by verb_phrase
this converter should process the adv after the verb
:param sentence:
:return:
"""
pattern = DependencyGraph()
# TODO: it seems that in UD labeling, adv is used instead of adj for noun
modified_node = DependencyGraphNode()
modifier_node = DependencyGraphNode()
pattern.add_nodes([modified_node, modifier_node])
pattern.add_dependency(modified_node, modifier_node, r'\bobl')
for match in dep_graph.match(pattern):
dep_modified_node = match[modified_node]
dep_modifier_node = match[modifier_node]
if oia_graph.has_relation(dep_modified_node,
dep_modifier_node,
direct_link=False):
continue
oia_modified_node = oia_graph.add_words(dep_modified_node.position)
oia_modifier_node = oia_graph.add_words(dep_modifier_node.position)
oia_graph.add_mod(oia_modifier_node, oia_modified_node)
def ccomp_mark_sconj(dep_graph: DependencyGraph):
"""
See them as they are
:param dep_graph:
:param oia_graph:
:return:
"""
pattern = DependencyGraph()
pred1_node = pattern.create_node(UPOS="VERB")
pred2_node = pattern.create_node()
sconj_node = pattern.create_node(UPOS="SCONJ")
pattern.add_dependency(pred1_node, pred2_node, r'ccomp')
pattern.add_dependency(pred2_node, sconj_node, 'mark')
for match in list(dep_graph.match(pattern)):
dep_pred1_node = match[pred1_node]
dep_pred2_node = match[pred2_node]
dep_sconj_node = match[sconj_node]
if dep_sconj_node.LEMMA == "as":
dep_graph.remove_dependency(dep_pred2_node, dep_sconj_node)
new_verb = [dep_pred1_node, "{1}", dep_sconj_node, "{2}"]
new_verb_node = merge_dep_nodes(new_verb,
UPOS=dep_pred1_node.UPOS,
LOC=dep_pred1_node.LOC)
# print("Noun detected", noun_node.ID)
dep_graph.replace_nodes(new_verb, new_verb_node)
def appositive_phrase(dep_graph, oia_graph, context: UD2OIAContext):
"""
##### Apposition: Trump, president of US, came #####
:param sentence:
:return:
"""
pattern = DependencyGraph()
subj_node = DependencyGraphNode()
appos_node = DependencyGraphNode()
pattern.add_nodes([subj_node, appos_node])
pattern.add_dependency(subj_node, appos_node, r'\w*appos\w*')
for match in dep_graph.match(pattern):
dep_subj_node = match[subj_node]
dep_appos_node = match[appos_node]
oia_appos_node = oia_graph.add_words(dep_appos_node.position)
oia_subj_node = oia_graph.add_words(dep_subj_node.position)
if oia_appos_node and oia_subj_node:
pred_node = oia_graph.add_aux(label="APPOS")
oia_graph.add_argument(pred_node, oia_subj_node, 1)
oia_graph.add_argument(pred_node, oia_appos_node, 2)
def adv_adj_modifier(dep_graph, oia_graph, context: UD2OIAContext):
"""
the adv before the verb should be processed by verb_phrase
this converter should process the adv after the verb
:param sentence:
:return:
"""
pattern = DependencyGraph()
# TODO: it seems that in UD labeling, adv is used instead of adj for noun
adj_node = DependencyGraphNode(UPOS="ADJ")
adv_node = DependencyGraphNode(UPOS="ADV|X|NOUN")
pattern.add_nodes([adj_node, adv_node])
pattern.add_dependency(adj_node, adv_node, r'advmod')
for match in dep_graph.match(pattern):
dep_adj_node = match[adj_node]
dep_adv_node = match[adv_node]
if oia_graph.has_relation(dep_adj_node, dep_adv_node):
continue
oia_adj_node = oia_graph.add_words(dep_adj_node.position)
oia_adv_node = oia_graph.add_words(dep_adv_node.position)
oia_graph.add_mod(oia_adv_node, oia_adj_node)
def part(dep_graph: DependencyGraph):
"""
:param dep_graph:
:return:
"""
pattern = DependencyGraph()
parent_node = pattern.create_node(UPOS="AUX|VERB")
part_node = pattern.create_node(UPOS="PART")
pattern.add_dependency(parent_node, part_node, r'advmod')
for match in list(dep_graph.match(pattern)):
dep_parent_node = match[parent_node]
dep_part_node = match[part_node]
new_node_list = [dep_parent_node, dep_part_node]
new_node_list.sort(key=lambda n: n.LOC)
new_node = merge_dep_nodes(new_node_list,
UPOS=dep_parent_node.UPOS,
LOC=dep_parent_node.LOC,
FEATS=dep_parent_node.FEATS
)
dep_graph.replace_nodes(new_node_list, new_node)
def whose_noun(dep_graph: DependencyGraph):
"""
:param dep_graph:
:return:
"""
pattern = DependencyGraph()
noun_node = pattern.create_node(UPOS="NOUN|PROPN|PRON|X|NUM|SYM")
owner_node = pattern.create_node()
whose_node = pattern.create_node(LEMMA="whose")
pattern.add_dependency(noun_node, owner_node, "nmod:poss")
pattern.add_dependency(owner_node, whose_node, "ref")
whose_noun_phrase = []
for match in dep_graph.match(pattern):
dep_owner_node = match[owner_node]
dep_noun_node = match[noun_node]
dep_whose_node = match[whose_node]
whose_noun_phrase.append(
(dep_owner_node, dep_whose_node, dep_noun_node))
for owner, whose, noun in whose_noun_phrase:
noun_node = merge_dep_nodes([whose, noun],
UPOS=noun.UPOS,
LOC=noun.LOC)
# print("Noun detected", noun_node.ID)
dep_graph.remove_dependency(owner_node, whose)
dep_graph.remove_dependency(noun, owner_node, "nmod:poss")
dep_graph.replace_nodes([whose, noun], noun_node)
def negation(dep_graph, oia_graph, context: UD2OIAContext):
"""
#################### Negation ########################
:param dep_graph:
:param oia_graph:
:return:
"""
pattern = DependencyGraph()
not_node = DependencyGraphNode(LEMMA="not")
parent_node = DependencyGraphNode()
pattern.add_nodes([not_node, parent_node])
pattern.add_dependency(parent_node, not_node, r'\w*')
for match in dep_graph.match(pattern):
dep_not_node = match[not_node]
dep_parent_node = match[parent_node]
oia_pred_node = oia_graph.add_aux(label="SCOPE")
oia_not_node = oia_graph.add_words(dep_not_node.position)
oia_parent_node = oia_graph.add_words(dep_parent_node.position)
oia_graph.add_argument(oia_pred_node, oia_not_node, 1)
oia_graph.add_argument(oia_pred_node, oia_parent_node, 1)
def there_be_verb_phrase(dep_graph):
"""
:param dep_graph:
:param oia_graph:
:return:
"""
pattern = DependencyGraph()
there_node = pattern.create_node(FORM=r'there|There')
be_node = pattern.create_node()
pattern.add_dependency(be_node, there_node, r'\w*expl\w*')
verb_phrases = []
for match in dep_graph.match(pattern):
dep_there_node = match[there_node]
dep_be_node = match[be_node]
if not "be" in dep_be_node.LEMMA.split(" "):
continue
pred = [dep_there_node, dep_be_node]
head = dep_be_node
verb_phrases.append((pred, head))
for verbs, root in verb_phrases:
verb_node = merge_dep_nodes(verbs, UPOS="VERB", LOC=root.LOC)
dep_graph.replace_nodes(verbs, verb_node)
def separated_asas(dep_graph: DependencyGraph):
"""
##### Equality comparison #####
##### A is as X a C as B #####
##### the first 'as' is always the advmod of a following element, X, which is within the range of as... as #####
##### the second 'as' is always the dependent of B #####
##### B sometimes depends on the first 'as', sometimes dependts on X #####
##### Sometimes X has a head that is also within the range of as...as #####
:param dep_graph:
:param oia_graph:
:return:
"""
pattern = DependencyGraph()
adj_node = DependencyGraphNode(UPOS="ADJ")
noun_node = DependencyGraphNode(UPOS="NOUN")
as1_node = DependencyGraphNode(FORM="as")
as2_node = DependencyGraphNode(FORM="as")
obj_node = DependencyGraphNode()
pattern.add_nodes([noun_node, adj_node, as1_node, as2_node, obj_node])
pattern.add_dependency(noun_node, adj_node, r'amod')
pattern.add_dependency(adj_node, as1_node, r'\w*advmod\w*')
pattern.add_dependency(as1_node, obj_node, r'\w*advcl:as\w*')
pattern.add_dependency(obj_node, as2_node, r'mark')
as_as_pred = []
for match in dep_graph.match(pattern):
dep_noun_node = match[noun_node]
dep_adj_node = match[adj_node]
dep_as1_node = match[as1_node]
dep_as2_node = match[as2_node]
dep_obj_node = match[obj_node]
if dep_as1_node.LOC < dep_adj_node.LOC < dep_noun_node.LOC < dep_as2_node.LOC < dep_obj_node.LOC:
pred = [
node for node in dep_graph.nodes()
if dep_as1_node.LOC <= node.LOC <= dep_adj_node.LOC
]
pred.append(dep_as2_node)
pred.sort(key=lambda x: x.LOC)
head = dep_adj_node
asas_node = merge_dep_nodes(pred, UPOS="ADJ", LOC=dep_as2_node.LOC)
as_as_pred.append(
(pred, head, asas_node, dep_noun_node, dep_obj_node))
for pred, head, asas_node, dep_noun_node, dep_obj_node in as_as_pred:
dep_graph.replace_nodes(pred, asas_node)
dep_graph.remove_dependency(asas_node, dep_obj_node)
dep_graph.remove_dependency(dep_noun_node, asas_node)
dep_graph.add_dependency(dep_noun_node, dep_obj_node,
"acl:" + asas_node.FORM)
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 process_head_conj(dep_graph: DependencyGraph):
"""
:param dep_graph:
:return:
"""
first_word = dep_graph.get_node_by_loc(0)
if first_word and first_word.LEMMA in {"and", "but"}:
cc_parents = [n for n, l in dep_graph.parents(first_word) if l == "cc"]
for p in cc_parents:
dep_graph.remove_dependency(p, first_word)
dep_graph.add_dependency(first_word, p, "arg_conj:1")
def adverbial_clause(dep_graph: DependencyGraph, oia_graph: OIAGraph,
context: UD2OIAContext):
"""
Adverbial Clause
##### run in order to catch it. advcl with mark (in order to) #####
##### he worked hard, replacing his feud. advcl without mark #####
:param dep_graph:
:param oia_graph:
:return:
"""
pattern = DependencyGraph()
verb_node = pattern.create_node()
modifier_node = pattern.create_node()
pattern.add_dependency(verb_node, modifier_node, "advcl")
for match in list(dep_graph.match(pattern)):
dep_verb_node = match[verb_node]
dep_modifier_node = match[modifier_node]
if context.is_processed(dep_verb_node, dep_modifier_node):
continue
oia_verb_node = oia_graph.add_words(dep_verb_node.position)
oia_modifier_node = oia_graph.add_words(dep_modifier_node.position)
logger.debug("adverbial clause: verb={0}, modifier={1}".format(
dep_verb_node.position, dep_modifier_node.position))
if oia_graph.has_relation(oia_verb_node, oia_modifier_node):
continue
mark = list(
dep_graph.children(dep_modifier_node,
filter=lambda n, rel: "mark" in rel))
if mark:
mark, rel = mark[0]
pred_node = oia_graph.add_words(mark.position)
if pred_node is None:
continue
if mark.LEMMA in CONJUNCTION_WORDS[language]:
continue
oia_graph.add_argument(pred_node, oia_verb_node, 1, mod=True)
oia_graph.add_argument(pred_node, oia_modifier_node, 2)
else:
oia_graph.add_mod(oia_modifier_node, oia_verb_node)
def complete_missing_case_mark(dep_graph: DependencyGraph, root, root_parents,
parallel_components, relation_to_conj,
case_marks):
"""
:param dep_graph:
:param parallel_components:
:return:
"""
parallel_components.sort(key=lambda x: x.LOC)
for parent in root_parents:
# ic(str(root))
# ic(str(parent))
# ic(relation_to_conj)
prefix, shared_prefix, required_mark = relation_to_conj[parent.ID]
if not required_mark:
continue
for index, (node,
mark) in enumerate(zip(parallel_components,
required_mark)):
if mark is None:
continue
is_exist = any(
mark == child.LEMMA or mark in child.LEMMA.split(" ")
for child, l in dep_graph.children(node))
if is_exist:
continue
found_mark = find_mark(case_marks, parallel_components[:index],
mark)
if found_mark:
mark_node, rel = found_mark
dup_case_mark = dep_graph.create_node(FORM=mark_node.FORM,
LEMMA=mark_node.LEMMA,
UPOS=mark_node.UPOS,
LOC=mark_node.LOC)
dup_case_mark.aux = True
dep_graph.add_dependency(node, dup_case_mark, rel)
else:
logger.warning("cannot find the mark, just add the relation")
def multi_words_mark(dep_graph: DependencyGraph):
"""
arise on to
the "on to" should be combined
:param dep_graph:
:param oia_graph:
:return:
"""
# print('multi_words_mark')
mark_phrases = []
for node in dep_graph.nodes():
marks = []
for n, l in dep_graph.children(node, filter=lambda n, l: "mark" in l):
marks.extend(dep_graph.offsprings(n))
if not marks:
continue
# print('multi_words_mark marks:', marks)
if len(marks) > 1:
if any([x.UPOS in {"NOUN", "NUM", "VERB", "ADJ", "ADV", "PRON"} for x in marks]):
continue
marks.sort(key=lambda n: n.LOC)
mark_phrases.append((node, marks))
for node, marks in mark_phrases:
# print('multi_words_mark marks:', marks)
if not all([dep_graph.get_node(x.ID) for x in marks]):
continue
mark_min_loc = marks[0].LOC
mark_max_loc = marks[-1].LOC
marks = [n for n in dep_graph.nodes() if mark_min_loc <= n.LOC <= mark_max_loc]
marks.sort(key=lambda n: n.LOC)
if any([x.UPOS in NOUN_UPOS for x in marks]):
continue
# print('marks:')
# for nnnn in marks:
# print(nnnn)
new_mark_node = merge_dep_nodes(marks,
UPOS=marks[0].UPOS,
LOC=marks[0].LOC
)
for mark in marks:
dep_graph.remove_dependency(node, mark)
dep_graph.replace_nodes(marks, new_mark_node)
dep_graph.add_dependency(node, new_mark_node, "mark")
def it_verb_clause(dep_graph: DependencyGraph, oia_graph: OIAGraph,
context: UD2OIAContext):
"""
##### Expletive #####
##### it is xxx to do #####
##### #####
:param dep_graph:
:param oia_graph:
:return:
"""
pattern = DependencyGraph()
it_node = pattern.create_node(LEMMA="it")
verb_node = pattern.create_node(UPOS="VERB")
subj_node = pattern.create_node(UPOS="NOUN|PRON|PROPN|VERB")
pattern.add_dependency(verb_node, it_node, r'expl')
pattern.add_dependency(verb_node, subj_node, r'nsubj|csubj')
for match in dep_graph.match(pattern):
dep_verb_node, dep_it_node, dep_subj_node = \
[match[x] for x in [verb_node, it_node, subj_node]]
if context.is_processed(dep_verb_node, dep_it_node):
continue
oia_it_node = oia_graph.add_words(dep_it_node.position)
oia_subj_node = oia_graph.add_words(dep_subj_node.position)
# oia_that_node = oia_graph.add_word_with_head(dep_that_node)
oia_verb_node = oia_graph.add_words(dep_verb_node.position)
if dep_it_node.LOC < dep_subj_node.LOC:
# it VERB subj that ...
oia_graph.add_argument(oia_verb_node, oia_it_node, 1)
oia_graph.add_argument(oia_verb_node, oia_subj_node, 1)
oia_graph.add_ref(oia_it_node, oia_subj_node)
else:
# subj VERB it that ...
oia_graph.add_argument(oia_verb_node, oia_it_node, 2)
oia_graph.add_argument(oia_verb_node, oia_subj_node, 2)
oia_graph.add_ref(oia_it_node, oia_subj_node)
# dep_graph.remove_dependency(dep_verb_node, dep_subj_node)
context.processed(dep_verb_node, dep_it_node)
context.processed(dep_verb_node, dep_subj_node)
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 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 multi_words_cc(dep_graph: DependencyGraph):
"""
arise on to
the "on to" should be combined
:param dep_graph:
:param oia_graph:
:return:
"""
mark_phrases = []
for node in dep_graph.nodes():
marks = []
for n, l in dep_graph.children(node, filter=lambda n, l: "cc" == l):
marks.extend(dep_graph.offsprings(n))
if not marks:
continue
if len(marks) > 1:
if any([x.UPOS in {"NOUN", "NUM", "VERB"} for x in marks]):
continue
marks.sort(key=lambda n: n.LOC)
mark_phrases.append((node, marks))
for node, marks in mark_phrases:
mark_min_loc = marks[0].LOC
mark_max_loc = marks[-1].LOC
marks = [n for n in dep_graph.nodes() if mark_min_loc <= n.LOC <= mark_max_loc]
if any([x.UPOS in {"NOUN", "NUM", "VERB"} for x in marks]):
continue
if not all([dep_graph.get_node(x.ID) for x in marks]):
continue
new_mark_node = merge_dep_nodes(marks,
UPOS=marks[0].UPOS,
LOC=marks[0].LOC
)
dep_graph.replace_nodes(marks, new_mark_node)
for mark in marks:
dep_graph.remove_dependency(node, mark)
if dep_graph.get_node(node.ID):
dep_graph.add_dependency(node, new_mark_node, "cc")
def xcomp_verb(dep_graph: DependencyGraph):
"""
:param dep_graph:
:return:
"""
pattern = DependencyGraph()
pred_node = pattern.create_node()
xcomp_verb_node = pattern.create_node(UPOS="VERB|AUX")
xcomp_mark_node = pattern.create_node(UPOS="PART")
pattern.add_dependency(pred_node, xcomp_verb_node, "xcomp")
pattern.add_dependency(xcomp_verb_node, xcomp_mark_node, "mark")
for match in list(dep_graph.match(pattern)):
dep_pred_node = match[pred_node]
dep_xcomp_verb_node = match[xcomp_verb_node]
dep_xcomp_mark_node = match[xcomp_mark_node]
if dep_xcomp_mark_node.LEMMA != "to":
# print('--------------------------LEMMA: ',dep_xcomp_mark_node.LEMMA)
# raise Exception("Unexpected Situation: xcomp mark != to let's throw out to see what happens")
continue
if dep_xcomp_mark_node.LOC > dep_xcomp_verb_node.LOC:
raise Exception(
"Unexpected Situation: xcomp mark after the xcomp verb")
pred_nodes = list(
dep_graph.parents(dep_xcomp_verb_node,
filter=lambda n, l: "xcomp" in l))
if len(pred_nodes) > 1:
raise Exception(
"Unexpected Situation: Multiple xcomp parents found")
new_verb_phrase = [dep_xcomp_mark_node, dep_xcomp_verb_node]
dep_new_verb = merge_dep_nodes(new_verb_phrase,
UPOS="VERB",
LOC=dep_xcomp_verb_node.LOC)
dep_graph.replace_nodes(new_verb_phrase, dep_new_verb)
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 such_that(dep_graph: DependencyGraph):
"""
##### such a high price that
:param dep_graph:
:param oia_graph:
:return:
"""
pattern = DependencyGraph()
noun_node = DependencyGraphNode(UPOS="NOUN")
such_node = DependencyGraphNode(FORM="such")
clause_pred_node = DependencyGraphNode(UPOS="VERB")
that_node = DependencyGraphNode(FORM="that")
pattern.add_nodes([noun_node, such_node, clause_pred_node, that_node])
pattern.add_dependency(noun_node, such_node, r'det:predet')
pattern.add_dependency(such_node, clause_pred_node, r'advcl:that')
pattern.add_dependency(clause_pred_node, that_node, r'mark')
such_that_pred = []
for match in dep_graph.match(pattern):
dep_noun_node = match[noun_node]
dep_such_node = match[such_node]
dep_clause_pred_node = match[clause_pred_node]
dep_that_node = match[that_node]
if dep_such_node.LOC < dep_noun_node.LOC < dep_that_node.LOC < dep_clause_pred_node.LOC:
such_that_pred.append((dep_noun_node, dep_such_node,
dep_clause_pred_node, dep_that_node))
for dep_noun_node, dep_such_node, dep_clause_pred_node, dep_that_node in such_that_pred:
nodes = [dep_such_node, dep_that_node]
such_that_pred = merge_dep_nodes(nodes,
UPOS="SCONJ",
LOC=dep_that_node.LOC)
dep_graph.add_node(such_that_pred)
dep_graph.add_dependency(dep_noun_node, dep_clause_pred_node,
"advcl:" + such_that_pred.FORM)
dep_graph.add_dependency(dep_clause_pred_node, such_that_pred, "mark")
dep_graph.remove_node(dep_such_node)
dep_graph.remove_node(dep_that_node)
def adv_question(dep_graph: DependencyGraph, oia_graph: OIAGraph,
context: UD2OIAContext):
"""
:param dep_graph:
:param oia_graph:
:return:
"""
pattern = DependencyGraph()
question_node = pattern.create_node(
UPOS="ADV|ADJ",
LEMMA=r"(\bhow\b|\bwhat\b|\bwhere\b|\bwhen\b|why\b)\w*")
verb_node = pattern.create_node(UPOS="VERB|AUX")
# subj_node = pattern.create_node()
pattern.add_dependency(verb_node, question_node, "advmod|amod")
# pattern.add_dependency(verb_node, subj_node, r"\w*subj")
for match in list(dep_graph.match(pattern)):
dep_question_node, dep_verb_node = \
[match[x] for x in (question_node, verb_node)]
# if not dep_question_node.LOC < dep_subj_node.LOC:
# # not a question
# continue
oia_question_node = oia_graph.add_words(dep_question_node.position)
oia_verb_node = oia_graph.add_words(dep_verb_node.position)
oia_graph.remove_relation(oia_verb_node, oia_question_node)
for parent, rel in list(oia_graph.parents(oia_verb_node)):
if rel.mod:
continue
oia_graph.remove_relation(parent, oia_verb_node)
oia_graph.add_relation(parent, oia_question_node, rel)
oia_graph.add_function(oia_question_node, oia_verb_node)
def det_of_noun(dep_graph: DependencyGraph):
"""
any/some/all of noun
:param dep_graph:
:return:
"""
pattern = DependencyGraph()
det_node = pattern.create_node(UPOS="DET")
of_node = pattern.create_node(LEMMA="of")
noun2_node = pattern.create_node(UPOS="NOUN|PROPN|PRON|X|NUM|SYM")
pattern.add_dependency(det_node, noun2_node, "nmod:of")
pattern.add_dependency(noun2_node, of_node, "case")
for match in list(dep_graph.match(pattern)):
dep_det_node = match[det_node]
dep_noun2_node = match[noun2_node]
dep_of_node = match[of_node]
if not all([dep_det_node, dep_noun2_node, dep_of_node]):
# processed by others
continue
if isinstance(dep_noun2_node,
DependencyGraphSuperNode) and dep_noun2_node.is_conj:
continue
dep_noun2_parents = [
parent for parent, rel in dep_graph.parents(dep_noun2_node)
]
if len(dep_noun2_parents) == 1:
assert dep_noun2_parents[0] == dep_det_node
new_noun_nodes = [dep_det_node, dep_of_node, dep_noun2_node]
new_noun = merge_dep_nodes(new_noun_nodes,
UPOS=dep_det_node.UPOS,
FEATS=dep_det_node.FEATS,
LOC=dep_det_node.LOC)
dep_graph.replace_nodes(new_noun_nodes, new_noun)
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 be_not_phrase(dep_graph: DependencyGraph):
"""TODO: add doc string
"""
pattern = DependencyGraph()
be_node = pattern.create_node() # contain the be verb
obj_node = pattern.create_node()
# not_node = pattern.create_node(UPOS="PART")
not_node = pattern.create_node()
pattern.add_node(be_node)
pattern.add_node(obj_node)
pattern.add_node(not_node)
pattern.add_dependency(be_node, obj_node, r'\w*obj\w*')
pattern.add_dependency(obj_node, not_node, r'\w*advmod\w*')
be_not = []
for match in dep_graph.match(pattern):
# print("be_not_phrase match !!!!!!!!!!!!!!")
dep_be_node = match[be_node]
dep_obj_node = match[obj_node]
dep_not_node = match[not_node]
if not "be" in dep_be_node.LEMMA.split(" "):
continue
if not "not" in dep_not_node.LEMMA.split(" "):
continue
if (dep_not_node.LOC > dep_obj_node.LOC) or (dep_be_node.LOC >
dep_not_node.LOC):
continue
be_not.append((dep_be_node, dep_obj_node, dep_not_node))
for dep_be_node, dep_obj_node, dep_not_node in be_not:
dep_graph.remove_dependency(dep_obj_node, dep_not_node, 'advmod')
verb_node = merge_dep_nodes((dep_be_node, dep_not_node),
UPOS=dep_be_node.UPOS,
LOC=dep_be_node.LOC)
dep_graph.replace_nodes([dep_be_node, dep_not_node], verb_node)
def adnominal_clause_mark(dep_graph, oia_graph, context: UD2OIAContext):
"""
#### Adnominal clause with marker #####
#### the way to go / the proof that doing this is helpful #####
:param sentence:
:return:
"""
pattern = DependencyGraph()
a = pattern.create_node()
b = pattern.create_node()
mark = pattern.create_node()
pattern.add_dependency(a, b, r'.*acl:.*')
pattern.add_dependency(b, mark, "mark")
for match in dep_graph.match(pattern):
dep_a, dep_b, dep_mark = [match[x] for x in [a, b, mark]]
if oia_graph.has_relation(dep_a, dep_b, direct_link=False):
continue
dep = dep_graph.get_dependency(dep_a, dep_b)
for rel in dep:
if "acl:" in rel:
# print('-----------------rel: ',rel)
prefix, value = rel.split(":")
if (value.replace("_", " ") in {dep_mark.FORM, dep_mark.LEMMA}
or value in dep_mark.FORM.split(" ")
or value in dep_mark.LEMMA.split(" ")):
a_node, b_node, mark_node = [
oia_graph.add_words(x.position)
for x in [dep_a, dep_b, dep_mark]
]
# TODO: decide whether function or argument
oia_graph.add_argument(mark_node, a_node, 1, mod=True)
oia_graph.add_argument(mark_node, b_node, 2)
context.processed(dep_a, dep_b)
context.processed(dep_b, dep_mark)
def aclwhose(dep_graph, oia_graph, context: UD2OIAContext):
"""
#### the person whose/who's cat is cute
#### @return a list of four-tuple (noun, whose/who's, possessee, aclmodifier)
:param sentence:
:return:
"""
pattern = DependencyGraph()
a = DependencyGraphNode() # person
b = DependencyGraphNode(FEATS={"PronType": "Int"}) # whose
c = DependencyGraphNode() # cat
d = DependencyGraphNode() # cute
pattern.add_nodes([a, b, c, d])
pattern.add_dependency(a, d, r'.*acl:relcl.*')
pattern.add_dependency(d, c, r'.*nsubj|obj|iobj.*')
pattern.add_dependency(c, b, r'.*nmod:poss.*')
# pattern.add_dependency(b, a, r'.*ref.*')
for match in dep_graph.match(pattern):
dep_a, dep_b, dep_c, dep_d = [match[x] for x in [a, b, c, d]]
a_node, b_node, c_node, d_node = [
oia_graph.add_words(x.position)
for x in [dep_a, dep_b, dep_c, dep_d]
]
oia_graph.add_function(d_node, a_node)
oia_graph.add_function(b_node, c_node)
oia_graph.add_ref(b_node, a_node)
