def eval(self, gold: Discourse, parse: Discourse):
if self.binarize:
gold = gold.binarize(left_heavy=self.left_heavy)
parse = parse.binarize(left_heavy=self.left_heavy)
self._eval_sent(gold, parse)
self._eval_edu(gold, parse)
self._eval_span(gold, parse)
self._eval_nuclear(gold, parse)
self._eval_relation(gold, parse)
self._eval_nuclear_classes(gold, parse)
self._eval_relation_classes(gold, parse)
def cut(self, label, text, start=0, end=-1, info=None):
if end < 0:
end = len(text)
sentences = []
edus = []
try:
for sentence in self.cut_sent(text, start, end):
sentences.append(sentence)
edus.extend(self.cut_edu(sentence))
except SentenceParseError as e:
logger.error("during segmenting %s, %s" % (label, e))
raise SegmentError("error segmenting %s" % label, label, text,
start, end, info)
return Discourse(label, text, (start, end), edus, sentences)
def sr_oracle(discorse: Discourse):
for node in discorse.binarize().traverse():
if isinstance(node, EDU):
yield SRTransition.SHIFT, None
elif isinstance(node, RelationNode):
yield SRTransition.REDUCE, node.nuclear
def _dom2discourse(p, label, info):
raw = p.find("RAW") # type: ElementTree.Element
_text = raw.get("Sentence")
sentspans = CDTB._xml2pos(raw.get("SentencePosition"))
_sentences = []
_offset = 0
for s, e in sentspans:
_sentences.append(_text[s + _offset:e + _offset])
_offset += 1
text = ''.join(_sentences)
sids = raw.get("SID").split("|") if raw.get("SID") else []
sentences = []
for i, span in enumerate(sentspans):
sid = sids[i] if sids else None
sentences.append(Sentence(span, text[slice(*span)], sid=sid))
eduspans = CDTB._xml2pos(raw.get("EduPosition"))
edus = [EDU(span, text[slice(*span)]) for span in eduspans]
del _text, _sentences, sids, sentspans, eduspans
discourse_span = CDTB._xml2pos(raw.get("AnnotatedPosition"))[0]
discourse = Discourse(label, text, discourse_span, edus, sentences,
info)
relations = {}
for r in p.find("Relations"): # type: ElementTree.Element
rid = r.get("ID")
prid = r.get("ParentId")
parent = relations[prid] if prid in relations else None
rspans = CDTB._xml2pos(r.get("SentencePosition"))
rspan = rspans[0][0], rspans[-1][1]
nuclear = nuclear_map[r.get("Center")] if r.get("Center") else None
if not r.get("ConnectiveType"):
relation_explicit = None
elif r.get("ConnectiveType") == "显式关系":
relation_explicit = True
else:
relation_explicit = False
if relation_explicit:
relation_connective = r.get("Connective").split("…")
relation_connectivespan = CDTB._xml2pos(
r.get("ConnectivePosition"))
else:
relation_connective = None
relation_connectivespan = None
relation_finetype = r.get("RelationType") or None
if "CoarseRelationType" in r.attrib and r.get(
"CoarseRelationType"):
relation_coarsetype = r.get("CoarseRelationType")
elif relation_finetype:
relation_coarsetype = coarsemap[relation_finetype]
else:
relation_coarsetype = None
relation_type = Relation(explicit=relation_explicit,
connective=relation_connective,
connective_span=relation_connectivespan,
fine=relation_finetype,
coarse=relation_coarsetype)
relation = discourse.add_relation(rspan,
nuclear,
relation=relation_type,
parent=parent)
relations[rid] = relation
return discourse
def _discourse2dom(discourse: Discourse):
p = et.Element("P")
p.attrib["ID"] = str(discourse.label)
raw = et.Element("RAW")
sent_spans = [sentence.span for sentence in discourse.sentences]
if sent_spans[0][0] != 0:
sent_spans = [(0, sent_spans[0][0])] + sent_spans
if sent_spans[-1][1] != len(discourse.text):
sent_spans = sent_spans + [(sent_spans[-1][1], len(discourse.text))
]
raw.attrib["AnnotatedPosition"] = CDTB._pos2xml([discourse.span])
raw.attrib["SentencePosition"] = CDTB._pos2xml(sent_spans)
raw.attrib["SID"] = ""
edu_spans = [edu.span for edu in discourse.edus]
raw.attrib["EduPosition"] = CDTB._pos2xml(edu_spans)
raw.attrib["Sentence"] = '|'.join(
[discourse.text[slice(*span)] for span in sent_spans])
if discourse.complete():
root = discourse.tree()
root_id = discourse.index(root)
raw.attrib["ROOT"] = str(root_id)
else:
raw.attrib["ROOT"] = ""
relations = et.Element("Relations")
relations.attrib["Nodes"] = str(len(discourse.relations))
if discourse.complete():
relations.attrib["Height"] = str(discourse.tree().height() - 2)
else:
relations.attrib["Height"] = ""
for relation in discourse.relations:
r = et.Element("R")
r.attrib["ID"] = str(discourse.index(relation))
if relation.nuclear:
r.attrib["Center"] = nuclear_map_rev[relation.nuclear]
else:
r.attrib["Center"] = ""
relations.append(r)
child_ids = [
discourse.index(child) for child in relation
if isinstance(child, RelationNode)
]
child_spans = [child.span for child in relation]
r.attrib["ChildList"] = "|".join(map(str, child_ids))
r.attrib["SentencePosition"] = CDTB._pos2xml(child_spans)
if relation.parent():
r.attrib["ParentId"] = str(discourse.index(relation.parent()))
else:
r.attrib["ParentId"] = "-1"
r.attrib["Sentence"] = "|".join(
[discourse.text[slice(*span)] for span in child_spans])
# relation type
relation_type = relation.relation
if relation_type:
explicit, fine, coarse, connective, connective_span = relation_type.explicit, \
relation_type.fine, \
relation_type.coarse, \
relation_type.connective, \
relation_type.connective_span
else:
explicit, fine, coarse, connective, connective_span = None, None, None, None, None
if explicit is not None:
r.attrib["ConnectiveType"] = "显式关系" if explicit else "隐式关系"
else:
r.attrib["ConnectiveType"] = ""
r.attrib["RelationType"] = fine or ""
if coarse:
r.attrib["CoarseRelationType"] = coarse
elif fine:
r.attrib["CoarseRelationType"] = coarsemap[fine]
else:
r.attrib["CoarseRelationType"] = ""
r.attrib["Connective"] = "…".join(
connective) if connective and explicit else ""
r.attrib["ConnectivePosition"] = CDTB._pos2xml(
connective_span) if explicit and connective_span else ""
p.append(raw)
p.append(relations)
return p