コード例 #1
0
 def _parse_mention(self, mention: Dict, doc: Doc) -> Mention:
     # get mention
     main_posi_info = mention.get("posi")
     if main_posi_info is None:
         return None  # no posi info!!
     sid, widx, wlen = self._read_posi(main_posi_info)
     ret = Mention.create(doc.sents[sid], widx, wlen)
     # possible head span?
     head_posi_info = mention.get("head", {}).get("posi")
     if head_posi_info is not None:
         head_sid, head_widx, head_wlen = self._read_posi(head_posi_info)
         if head_sid != sid or not (head_widx>=widx and head_widx+head_wlen<=widx+wlen):
             zwarn(f"Error in head: {head_posi_info} vs. {main_posi_info}")
         else:  # make sure things are correct! otherwise simply discard!!
             ret.set_span(head_widx, head_wlen, hspan=True)
     return ret
コード例 #2
0
ファイル: frame.py プロジェクト: zzsfornlp/zmsp
 def _fake_discontinuous_ef(self, all_mentions: List):
     # todo(+N): for simplicity, special support for it!!
     # simply make the fake ones but no adding!!
     start_widx, end_widx = all_mentions[0].widx, all_mentions[0].wridx
     for m in all_mentions[1:]:
         if m.sent is all_mentions[0].sent:  # expand
             start_widx = min(start_widx, m.widx)
             end_widx = max(end_widx, m.wridx)
     fake_ef = Frame.create(mention=Mention.create(all_mentions[0].sent, start_widx, end_widx - start_widx))
     # --
     excluded_idxes = set(range(start_widx, end_widx))
     for m in all_mentions:
         for i in range(m.widx, m.wridx):
             if i in excluded_idxes:  # there can be repeated mentions
                 excluded_idxes.remove(i)
     fake_ef.mention.excluded_idxes = excluded_idxes  # todo(+N): ugly fix!!
     return fake_ef