def remove_trailing_periods(text, offset):
"""
:type text: str
:type offset: IntPair
"""
newtext = text
newoffset = IntPair(offset.first, offset.second)
chars = set(['.', ',', ':', ';', ')', '}', ']', '"', '\'', '?', '!'])
if text[-1] in chars:
i = 1
while text[-(i+1)] is ' ':
i += 1
newtext = text[0:-i]
newoffset.second = newoffset.second - i
return newtext, newoffset
def process_span_file(doc, filename):
"""Reads event annotation from filename, and add to doc
:type filename: str
:type doc: nlplingo.text.text_theory.Document
<Event type="CloseAccount">
CloseAccount 0 230
anchor 181 187
CloseAccount/Source 165 170
CloseAccount/Source 171 175
CloseAccount/Source 176 180
CloseAccount/Target 191 198
CloseAccount/Target 207 214
CloseAccount/Target 215 229
</Event>
"""
lines = []
""":type: list[str]"""
with codecs.open(filename, 'r', encoding='utf-8') as f:
for line in f:
lines.append(line.strip())
i = 0
while i < len(lines):
line = lines[i]
if line.startswith('<Event type='):
event_type = re.search(r' type="(.*?)"', line).group(1)
event_id = '{}.e-{}'.format(doc.docid, len(doc.events))
event = Event(event_id, event_type)
i += 1
line = lines[i]
while not line.startswith('</Event>'):
tokens = line.split()
info = tokens[0]
offset = IntPair(int(tokens[1]), int(tokens[2]))
if info == event_type or info == 'anchor' or '/' in info:
text = doc.get_text(offset.first, offset.second)
if text is None or text == '':
print('WARNING: skipping annotation span {} {}-{}'.format(doc.docid, offset.first, offset.second))
else:
# sometimes, the UI captures an extra trailing space. Check for that and adjust ending offset
if text[-1] == ' ':
text = text[0:-1]
offset.second = offset.second - 1
if info == event_type: # this is an event span
id = '{}.s-{}'.format(event_id, len(event.event_spans))
event.add_event_span(EventSpan(id, offset, text, event_type))
elif info == 'anchor': # anchor span
id = '{}.t-{}'.format(event_id, len(event.anchors))
#print('Spannotator, adding ANCHOR with text "{}"'.format(text))
newtext, newoffset = remove_trailing_periods(text, offset)
if text != newtext:
print('- revising anchor, text=[%s] offset=(%d,%d) newtext=[%s] newoffset=(%d,%d)' % (text, offset.first, offset.second, newtext, newoffset.first, newoffset.second))
event.add_anchor(Anchor(id, newoffset, newtext, event_type))
elif '/' in info: # argument span
em_id = 'm-{}-{}'.format(offset.first, offset.second)
newtext, newoffset = remove_trailing_periods(text, offset)
if text != newtext:
print('- revising argument, text=[%s] offset=(%d,%d) newtext=[%s] newoffset=(%d,%d)' % (text, offset.first, offset.second, newtext, newoffset.first, newoffset.second))
em = EntityMention(em_id, newoffset, newtext, 'dummy')
# we just use a dummy em first, for creating the EventArgument (notice that this em is not added to the doc)
# later, when we annotate sentence, we will find an actual EntityMention that is backed by tokens
# and use that to back the EventArgument
# Ref: text_theory.annotate_sentence_with_events()
arg_role = info[info.index('/') + 1:]
arg_id = '{}.t-{}'.format(event_id, len(event.arguments))
event.add_argument(EventArgument(arg_id, em, arg_role))
i += 1
line = lines[i]
doc.add_event(event)
i += 1
return doc