def style_convert_string(self, input_text):
""" For each word in input text, look up synonyms in the
author's thesaurus and probabilistically select a
replacement word. Write output to outfile. """
text = tokenize_string(input_text)
output = ""
tagged_tuples = nltk.pos_tag(text)
untagged_string = " ".join([tagged_tuple[0] for tagged_tuple in tagged_tuples])
for index, tagged_tuple in enumerate(tagged_tuples):
orig_word, temp_pos = tagged_tuple
word = orig_word.strip().lower()
was_title = orig_word.istitle() # "Title"
was_capitalized = orig_word.isupper() # "UPPER"
was_lower = orig_word.islower() # "lower"
# Don't replace determinants
if temp_pos == u'DT':
weighted_key = word
else:
if temp_pos in ['VBD', 'VBG', 'VBN', 'NNS', 'NNPS']:
synset = None
else:
# Replace word
# Converts penn tree bank pos tag to wordnet pos tag
wordnet_pos = reduce_pos_tagset(temp_pos)
if wordnet_pos:
synset = nltk_lesk(untagged_string, word, wordnet_pos)
else:
synset = nltk_lesk(untagged_string, word)
# Probabilistically choose a synonym in thesaurus[synset]
weighted_key = self._weighted_choice_lesk(str(synset), word)
# Match capitalization of original word
if was_title:
weighted_key = weighted_key.title()
elif was_capitalized:
weighted_key = weighted_key.upper()
elif not was_lower:
weighted_key = orig_word
# Add a space between words, no space for punctuation
if word not in string.punctuation and index != 0:
output += " "
output += weighted_key
return output
def lesk_builtin(wsd):
"""returns word sense for synset found using lesk's algorithm"""
synset = nltk_lesk(wsd.context, wsd.lemma)
if synset is not None:
return get_first_sense_key(synset)
else:
logger.debug('synset empty for {}'.format(wsd.lemma))
return (None)
def lesk(self, context, word):
from nltk.wsd import lesk as nltk_lesk
context_words = self.context2words(context)
return nltk_lesk(context_words, word, 'n')
def lesk(self, context, word):
from nltk.wsd import lesk as nltk_lesk
context_words = self.context2words(context)
return nltk_lesk(context_words, word, 'n')
def style_convert_lesk(self, infile_name, outfile_name):
""" For each word in input text, look up synonyms in the
author's thesaurus and probabilistically select a
replacement word. Write output to outfile. """
with open(infile_name, 'r') as infile, open(outfile_name, 'w') as outfile:
for line in infile:
# POS tag, and then lesk-ify the input,
# look it up in the thesauri
try:
line = line.decode('ascii', 'ignore')
except (UnicodeDecodeError, UnicodeEncodeError):
continue
line = tokenize_string(line)
tagged_tuples = nltk.pos_tag(line)
untagged_string = " ".join([tagged_tuple[0] for tagged_tuple in tagged_tuples])
for index, tagged_tuple in enumerate(tagged_tuples):
orig_word, temp_pos = tagged_tuple
word = orig_word.strip().lower()
was_title = orig_word.istitle() # "Title"
was_capitalized = orig_word.isupper() # "UPPER"
was_lower = orig_word.islower() # "lower"
# Don't replace determinants
if temp_pos == u'DT':
weighted_key = word
else:
# Skip past tense verbs and nouns for synsets
if temp_pos in ['VBD', 'VBG', 'VBN', 'NNS', 'NNPS']:
synset = None
else:
# Replace word
# Converts penn tree bank pos tag to wordnet pos tag
wordnet_pos = reduce_pos_tagset(temp_pos)
if wordnet_pos:
synset = nltk_lesk(untagged_string, word, wordnet_pos)
else:
synset = nltk_lesk(untagged_string, word)
# Probabilistically choose a synonym in thesaurus[synset]
# -> Interpolates to non-WordNet/Synset if synset doesn't exist
weighted_key = self._weighted_choice_lesk(str(synset), word)
# Match capitalization of original word
if was_title:
weighted_key = weighted_key.title()
elif was_capitalized:
weighted_key = weighted_key.upper()
elif not was_lower:
weighted_key = orig_word
# Add a space between words, no space for punctuation
if word not in string.punctuation and index != 0:
outfile.write(" ")
outfile.write(weighted_key)
outfile.write('\n')
return outfile_name
