def wordnet_get_gloss_byref(_ref):
"""
given a wordnet reference number returns term gloss
_ref is the wordnet referenace number (e.g friday%1:28:00::)
"""
if not _ref: # if ref is empty
return ''
try:
return wn.synset_from_sense_key(_ref).definition()
except (AttributeError,WordNetError,ValueError) as err:
return 'WN Error'
def get_gloss_relation(self, name):
synset = wn.synset(name)
synset_pos = synset.pos()
if synset_pos == "s":
synset_pos = "a"
synset_id = 'wn:{}{}'.format(str(synset.offset()).zfill(8), synset_pos)
related_synset_keys = self.sense.getGlossRelatedWordNetSynsetIds(
synset_id)
related_gloss_synsets = []
for synset_key in related_synset_keys:
try:
related_gloss_synset = wn.synset_from_sense_key(synset_key)
related_gloss_synsets.append(related_gloss_synset)
except:
continue
return related_gloss_synsets
def get_other_senses_byref(_ref,select_name=False):
"""
given a wordnet reference number returns homonyms list in the form of
other synsets
_ref is the wordnet reference number (e.g friday%1:28:00::)
"""
try:
ref_syn = wn.synset_from_sense_key(_ref)
ref_word = ref_syn.name().split('.')[0]
if not select_name:
return [syn for syn in wn.synsets(ref_word) if syn != ref_syn]
return [syn for syn in wn.synsets(ref_word) if (syn != ref_syn and ref_word in syn.name())]
except (AttributeError,WordNetError,KeyError,ValueError) as err:
return 'WN Error'
def create_hypernynm_gloss_data(config):
train_files = config['train_raw_files']
gloss_data, hyp_data = [], []
oversample_ratio = config['oversample_ratio']
triplet_data = {}
for file in train_files:
train_file = open(os.path.join(config['train_raw_dir'], file),
'r',
encoding='utf8')
for line in train_file.readlines()[1:]:
info = line.strip().split('\t')
target_id, label, sentence, gloss, sense_key = info[0], info[
1], info[2], info[3], info[4]
gloss_data.append([target_id, label, sentence, gloss, sense_key])
if sentence in triplet_data:
triplet_obj = triplet_data[sentence]
else:
triplet_obj = TripletSample(sentence)
triplet_data[sentence] = triplet_obj
if label == '1':
triplet_obj.positives.add(gloss)
else:
triplet_obj.negatives.add(gloss)
if label == '1':
for _ in range(oversample_ratio - 1):
gloss_data.append(
[target_id, label, sentence, gloss, sense_key])
try:
wn_synset = wn.synset_from_sense_key(sense_key)
if len(wn_synset.hypernyms()) > 0:
for hyp in wn_synset.hypernyms():
defs = hyp.definition()
word = hyp.lemmas()[0].name().replace('_', ' ')
new_gloss = word + ' : ' + defs
hyp_data.append(
[target_id, label, sentence, new_gloss, sense_key])
if label == '1':
for _ in range(oversample_ratio - 1):
hyp_data.append([
target_id, label, sentence, new_gloss,
sense_key
])
except:
continue
train_file.close()
return gloss_data, hyp_data, triplet_data
def accuracy(y, y_pred, by_synset=False):
n = len(y)
k = 0
for key in y.keys():
try:
senses = [
wn.synset_from_sense_key(sense) if by_synset else sense
for sense in y[key]
]
if y_pred[key].lemmas()[0].key() in senses:
k += 1
#else: print(y_pred[key], senses) # Why does wordnet.synset_from_sense_key return 'brazil_nut.n.02'?
# Are these labels all right?
except WordNetError:
pass # This should only occur for "budget"
# BAD PRACTICE
return k / n
def get_lemma_classifiers(instances,
keys,
classifier=MultinomialNB,
binary=False,
idf=True):
'''
:type instances: dict of WSDInstance
:type keys: dict of list
'''
classifiers = {}
data = {}
for k, inst in instances.items():
if inst.lemma not in data.keys():
data[inst.lemma] = [[], []] # [[context lists], [sense label]]
try:
for sense in keys[k]:
data[inst.lemma][0].append(inst.context)
data[inst.lemma][1].append(
wn.synset_from_sense_key(sense).name())
except WordNetError as e: # "budget" again.
pass
for k, d in data.items():
if not d[1]: continue # 'budget' strikes again
vctr = TfidfVectorizer(lowercase=False,
preprocessor=lambda x: x,
tokenizer=lambda x: x,
binary=binary,
use_idf=idf)
x = vctr.fit_transform(d[0])
clfr = classifier()
clfr.fit(x.toarray(), d[1])
if len(
clfr.classes_
) > 1: # Oh great... Should have used SemCor but don't have time.
classifiers[k] = {'vct': vctr, 'clf': clfr}
return classifiers
def get_synset_id_from_sense_key(sense_key):
return wn.synset_from_sense_key(sense_key).name()
from nltk.corpus import wordnet as wn
dataset = 'ALL'
if __name__ == "__main__":
with open(f'./data/mono/evaluation/{dataset}/{dataset}_n.gold.key.txt',
'w') as out:
for line in open(
f'./data/mono/evaluation/{dataset}/{dataset}.gold.key.txt'):
line = line.replace('\n', '')
id = line.split(' ')[0]
sense_key = line.split(' ')[1]
syn = wn.synset_from_sense_key(sense_key)
if syn.pos() != 'n':
continue
syn_id = syn.name()
out.write(f"{id} {syn_id}\n")
