def load(
self,
document: Document,
cases: List[str],
exophors: List[str],
coreference: bool,
bridging: bool,
relations: List[str],
kc: bool,
pas_targets: List[str],
tokenizer: BertTokenizer,
) -> None:
self.doc_id = document.doc_id
process_all = (kc is False) or (document.doc_id.split('-')[-1] == '00')
last_sent = document.sentences[-1] if len(document) > 0 else None
relax_exophors = {}
for exophor in exophors:
relax_exophors[exophor] = exophor
if exophor in ('不特定:人', '不特定:物', '不特定:状況'):
for n in '１２３４５６７８９':
relax_exophors[exophor + n] = exophor
dmid2arguments: Dict[int, Dict[str, List[BaseArgument]]] = {
pred.dmid: document.get_arguments(pred)
for pred in document.get_predicates()
}
head_dmids = []
for sentence in document:
process: bool = process_all or (sentence is last_sent)
head_dmids += [bp.dmid for bp in sentence.bps]
for bp in sentence.bps:
for mrph in bp.mrph_list():
self.words.append(mrph.midasi)
self.dtids.append(bp.dtid)
self.ddeps.append(
bp.parent.dtid if bp.parent is not None else -1)
arguments = OrderedDict((rel, []) for rel in relations)
arg_candidates = ment_candidates = []
if document.mrph2dmid[mrph] == bp.dmid and process is True:
if is_pas_target(bp,
verbal=('pred' in pas_targets),
nominal=('noun' in pas_targets)):
arg_candidates = [
x for x in head_dmids if x != bp.dmid
]
for case in cases:
dmid2args = {
dmid: arguments[case]
for dmid, arguments in
dmid2arguments.items()
}
arguments[case] = self._get_args(
bp.dmid, dmid2args, relax_exophors,
arg_candidates)
if bridging and is_bridging_target(bp):
arg_candidates = [
x for x in head_dmids if x != bp.dmid
]
dmid2args = {
dmid: arguments['ノ']
for dmid, arguments in dmid2arguments.items()
}
arguments['ノ'] = self._get_args(
bp.dmid, dmid2args, relax_exophors,
arg_candidates)
if coreference and is_coreference_target(bp):
ment_candidates = [
x for x in head_dmids if x < bp.dmid
] # do not solve cataphora
arguments['='] = self._get_mentions(
bp, document, relax_exophors, ment_candidates)
self.arguments_set.append(arguments)
self.arg_candidates_set.append(arg_candidates)
self.ment_candidates_set.append(ment_candidates)
def coverage(doc: Document) -> RetValue:
ret = RetValue()
for predicate in doc.get_predicates():
ex = Example(doc, predicate)
arguments = doc.get_arguments(predicate)
is_pred_gold = any(arguments[case] for case in PRED_CASES)