def translate_coref(infile, outfile, translated):
with open(infile, 'r') as f:
soup = bs4.BeautifulSoup(f, 'xml')
total = 0
skipped = 0
for mrk in soup.find_all('markable'):
total += 1
span_parts = []
for in_span in mrk['span'].split(','):
from_idx, to_idx = mmax.parse_span(in_span)
if from_idx not in translated:
print('Unaligned start word: ' + str(mrk), file=sys.stderr)
skipped += 1
continue
if from_idx == to_idx:
span = 'word_%d' % (translated[from_idx] + 1)
elif to_idx - 1 not in translated:
print('Unaligned end word: ' + str(mrk), file=sys.stderr)
skipped += 1
continue
else:
span = 'word_%d..word_%d' % tuple(
translated[i] + 1 for i in [from_idx, to_idx - 1])
span_parts.append(span)
mrk['span'] = ','.join(span_parts)
print('Skipped %d out of %d markables.' % (skipped, total),
file=sys.stderr)
with open(outfile, 'w') as f:
print(soup.prettify(), file=f)
def get_coref_chain_boundaries(mmax_dir, mmax_id):
with open(mmax.sentences_file(mmax_dir, mmax_id), 'r') as f:
s_soup = bs4.BeautifulSoup(f, 'xml')
sentence_id = {}
for mrk in s_soup.find_all('markable'):
for i in range(*mmax.parse_span(mrk['span'])):
sentence_id[i] = mrk['orderid']
with open(mmax.coref_file(mmax_dir, mmax_id), 'r') as f:
soup = bs4.BeautifulSoup(f, 'xml')
directory = {'__next__': 1}
boundaries = {}
clause_or_vp = set()
for mrk in soup.find_all('markable'):
if not mrk.has_attr('coref_class') or not mrk['coref_class'] or mrk['coref_class'] == 'empty':
continue
chain_idx = lookup_chain(directory, mrk['coref_class'])
if mrk['mention'] in ('clause', 'vp'):
clause_or_vp.add(chain_idx)
for s in mrk['span'].split(','):
start, end = mmax.parse_span(s)
if start == end - 1:
append(boundaries, start, ('(%d)', chain_idx))
else:
if sentence_id[start] != sentence_id[end - 1]:
print('%s: Skipped cross-sentence mention (%d): %s' % (mmax_id, end - start, str(mrk)),
file=sys.stderr)
else:
append(boundaries, start, ('(%d', chain_idx))
append(boundaries, end - 1, ('%d)', chain_idx))
str_boundaries = {}
for pos, chains in boundaries.items():
str_chains = [fmt % idx for fmt, idx in chains if idx not in clause_or_vp]
if len(str_chains):
str_boundaries[pos] = '|'.join(str_chains)
else:
str_boundaries[pos] = '-'
return str_boundaries
def compare_mmax(dir1, dir2, mmax_dir, mmax_id):
mmax_dir1 = os.path.join(dir1, mmax_dir)
mmax_dir2 = os.path.join(dir2, mmax_dir)
with open(mmax.words_file(mmax_dir1, mmax_id), 'r') as f:
words1 = [
w.string for w in bs4.BeautifulSoup(f, 'xml').find_all('word')
]
with open(mmax.words_file(mmax_dir2, mmax_id), 'r') as f:
words2 = [
w.string for w in bs4.BeautifulSoup(f, 'xml').find_all('word')
]
to_check = []
for level in ['sentence', 'coref']:
fname1 = '%s/Markables/%s_%s_level.xml' % (mmax_dir1, mmax_id, level)
fname2 = '%s/Markables/%s_%s_level.xml' % (mmax_dir2, mmax_id, level)
with open(fname1, 'r') as f:
markables1 = [
m for m in bs4.BeautifulSoup(f, 'xml').find_all('markable')
]
with open(fname2, 'r') as f:
markables2 = [
m for m in bs4.BeautifulSoup(f, 'xml').find_all('markable')
]
def sort_key(mrk):
return [mmax.parse_span(m) for m in mrk['span'].split(',')]
markables1.sort(key=sort_key)
markables2.sort(key=sort_key)
if len(markables1) != len(markables2):
print('%s/%s: Number of markables does not match (%d != %d)' %
(mmax_dir, mmax_id, len(markables1), len(markables2)),
file=sys.stderr)
continue
for mrk1, mrk2 in zip(markables1, markables2):
spans1 = mrk1['span'].split(',')
spans2 = mrk2['span'].split(',')
if len(spans1) != len(spans2):
print(
'%s/%s: Number of span components does not match (%s / %s)'
% (mmax_dir, mmax_id, mrk1['span'], mrk2['span']),
file=sys.stderr)
continue
for sp1, sp2 in zip(spans1, spans2):
s1, e1 = mmax.parse_span(sp1)
s2, e2 = mmax.parse_span(sp2)
txt1 = ''.join(words1[s1:e1])
txt2 = ''.join(words2[s2:e2])
# Quotes are changed by the tokeniser
txt1 = txt1.replace("``", '"').replace("''", '"')
txt2 = txt2.replace("``", '"').replace("''", '"')
if txt1 != txt2:
to_check.append((level, sp2))
if to_check:
print(mmax_dir, mmax_id, file=sys.stderr)
create_checks_level(mmax_dir2, mmax_id, to_check)
def sort_key(mrk):
return [mmax.parse_span(m) for m in mrk['span'].split(',')]
