def _read_entry(self, entry):
ds = entry["datasource"].lower()
if ds == 'wikimwe':
words = entry["lemmas"] # not actually always lemmatized
del entry["lemmas"]
entry["words"] = words
if ds in self.POS_2_PENN: # map POSes to Penn Treebank tagset
for i, p in enumerate(entry["poses"]):
info = self.POS_2_PENN[ds].get(p, p)
entry["poses"][i] = info if isinstance(
info, basestring) else (info.get(
entry["lemmas" if ds == 'baldwin vpc' else "words"][i])
or info[None])
if "lemmas" not in entry:
if 'lvc' in ds:
entry["lemmas"] = [
entry["verblemma"],
morph.stem(entry["noun"], 'NN')
]
else:
assert "words" in entry, entry
words = entry["words"]
poses = [None] * len(words)
if "poses" in entry and entry["poses"]:
assert ds in {'said', 'semcor', 'wikimwe'}, entry
poses = entry["poses"]
elif ds in {'phrases.net', "oyz's idioms"}:
pass
elif entry["label"].startswith(
'NNP') or entry["label"].startswith('NE:'):
poses = ['NNP'] * len(words)
entry["lemmas"] = [
morph.stem(w, p) for w, p in zip(words, poses)
]
try:
sig = tuple(l.lower() for l in entry["lemmas"]
if not l[0] == l[-1] == '_')
if not sig or sig[-1] == 'the' or not any(
l for l in sig if len(l) > 2):
return # probably garbage entry
if len(sig) > 1:
self._entries[sig] = entry
self._bylast[sig[-1]].add(sig)
except:
print(entry, file=sys.stderr)
raise
def _read_nonblank_line(self, ln, sent):
'''Tab-separated format:
word pos tag sentId
tag and sentId are optional.
'''
parts = ln[:-1].split('\t')[:4]
if len(parts) == 4:
token, pos, tag, sentId = parts
sent.sentId = sentId
elif len(parts) == 3:
token, pos, tag = parts
if not tag.strip():
tag = None
else:
token, pos = parts
tag = None
if tag is not None:
if self._labels is None:
pass
elif tag == '0' and self._legacy0:
assert 'O' in self._labels, self._labels
tag = 'O'
elif tag not in self._labels:
tag = 'O'
tag = uintern(unicode(tag))
pos = uintern(unicode(pos))
stemS = uintern(unicode(morph.stem(token, pos)))
sent.addToken(token=token, stem=stemS, pos=pos, goldTag=tag)
def _read_nonblank_line(self, ln, sent):
'''Tab-separated format:
word pos tag sentId
tag and sentId are optional.
'''
parts = ln[:-1].split('\t')[:4]
if len(parts)==4:
token, pos, tag, sentId = parts
sent.sentId = sentId
elif len(parts)==3:
token, pos, tag = parts
if not tag.strip():
tag = None
else:
token, pos = parts
tag = None
if tag is not None:
if self._labels is None:
pass
elif tag=='0' and self._legacy0:
assert 'O' in self._labels,self._labels
tag = 'O'
elif tag not in self._labels:
tag = 'O'
tag = uintern(unicode(tag))
pos = uintern(unicode(pos))
stemS = uintern(unicode(morph.stem(token,pos)))
sent.addToken(token=token, stem=stemS, pos=pos, goldTag=tag)
def _read_nonblank_line(self, ln, sent):
'''Tab-separated format:
offset word lemma POS tag parent strength label sentId
lemma will (for now) be ignored in favor of the automatic stemmer.
label may be the empty string; sentId is optional.
'''
parts = ln[:-1].split('\t')
if len(parts)==9:
offset, token, _, pos, tag, parent, strength, label, sentId = parts
sent.sentId = sentId
else:
offset, token, _, pos, tag, parent, strength, label = parts
offset = int(offset)
parent = int(parent)
assert len(sent)+1==offset
assert parent<offset
if tag is not None:
if self._labels is None:
pass
elif tag=='0' and self._legacy0:
assert 'O' in self._labels,self._labels
tag = 'O'
elif tag not in self._labels:
tag = 'O'
tag = uintern(unicode(tag))
pos = uintern(unicode(pos))
stemS = uintern(unicode(morph.stem(token,pos)))
sent.addToken(token=token, stem=stemS, pos=pos, goldTag=tag,
goldparent=int(parent), goldstrength=uintern(unicode(strength)),
goldlabel=uintern(unicode(label)))
def _read_entry(self, entry):
ds = entry["datasource"].lower()
if ds=='wikimwe':
words = entry["lemmas"] # not actually always lemmatized
del entry["lemmas"]
entry["words"] = words
if ds in self.POS_2_PENN: # map POSes to Penn Treebank tagset
for i,p in enumerate(entry["poses"]):
info = self.POS_2_PENN[ds].get(p,p)
entry["poses"][i] = info if isinstance(info,basestring) else (info.get(entry["lemmas" if ds=='baldwin vpc' else "words"][i]) or info[None])
if "lemmas" not in entry:
if 'lvc' in ds:
entry["lemmas"] = [entry["verblemma"], morph.stem(entry["noun"],'NN')]
else:
assert "words" in entry,entry
words = entry["words"]
poses = [None]*len(words)
if "poses" in entry and entry["poses"]:
assert ds in {'said','semcor','wikimwe'},entry
poses = entry["poses"]
elif ds in {'phrases.net', "oyz's idioms"}:
pass
elif entry["label"].startswith('NNP') or entry["label"].startswith('NE:'):
poses = ['NNP']*len(words)
entry["lemmas"] = [morph.stem(w,p) for w,p in zip(words,poses)]
try:
sig = tuple(l.lower() for l in entry["lemmas"] if not l[0]==l[-1]=='_')
if not sig or sig[-1]=='the' or not any(l for l in sig if len(l)>2):
return # probably garbage entry
if len(sig)>1:
self._entries[sig] = entry
self._bylast[sig[-1]].add(sig)
except:
print(entry, file=sys.stderr)
raise
def _read_nonblank_line(self, ln, sent):
'''Tab-separated format:
offset word lemma POS tag parent strength label sentId
lemma will (for now) be ignored in favor of the automatic stemmer.
label may be the empty string; sentId is optional.
'''
parts = ln[:-1].split('\t')
if len(parts) == 9:
offset, token, _, pos, tag, parent, strength, label, sentId = parts
sent.sentId = sentId
else:
offset, token, _, pos, tag, parent, strength, label = parts
offset = int(offset)
parent = int(parent)
assert len(sent) + 1 == offset
assert parent < offset
if tag is not None:
if self._labels is None:
pass
elif tag == '0' and self._legacy0:
assert 'O' in self._labels, self._labels
tag = 'O'
elif tag not in self._labels:
tag = 'O'
tag = uintern(unicode(tag))
pos = uintern(unicode(pos))
stemS = uintern(unicode(morph.stem(token, pos)))
sent.addToken(token=token,
stem=stemS,
pos=pos,
goldTag=tag,
goldparent=int(parent),
goldstrength=uintern(unicode(strength)),
goldlabel=uintern(unicode(label)))
