def initializeSoundexDict(freqDict):
soundexScore = Soundex().soundex
freqDictSoundex = {}
for c in freqDict:
freqDictSoundex[c] = soundexScore(c.capitalize())
return freqDictSoundex
def soundex(word):
word = word.capitalize()
sound_candidates = []
soundexScore = Soundex().soundex
wordScore = soundexScore(word)
for c in freqDict:
if (wordScore == freqDictSoundex[c]):
sound_candidates.append(c)
if len(sound_candidates) == 0:
sound_candidates.append(word)
words = []
for i in range(min(len(sound_candidates), candidates_nb)):
words.append(max(sound_candidates, key=P))
sound_candidates.remove(words[i])
return words
def __init__(self):
"""
Initialize necessary resources.
"""
self.dictionary_file = open(os.path.join(
os.path.dirname(__file__), 'data/ml_rootwords.txt'))
self.dictionary = self.dictionary_file.readlines()
self.dictionary_file.close()
try:
self.dictionary = marisa_trie.Trie([x.strip().decode('utf-8')
for x in self.dictionary])
except:
self.dictionary = marisa_trie.Trie(
[x.strip() for x in self.dictionary])
self.stemmer = Stemmer()
self.inflector = inflector.Inflector(lang='ml')
self.soundex = Soundex()
self.syllabalizer = Syllabalizer()
self.ngrammer = Ngram()
def has_player_name_fuzzy(current, edu):
"if the EDU has a word that sounds like a player name"
tokens = edu.tokens
soundex = lambda w: Soundex().soundex(w)
return has_one_of_words(current.players, tokens, norm=soundex)
class BaseMalayalam:
"""
Malayalam Spell Checker class.
"""
Suggestion = namedtuple('Suggestion', 'word sound lev jac weight tag_list')
def __init__(self):
"""
Initialize necessary resources.
"""
self.dictionary_file = open(os.path.join(
os.path.dirname(__file__), 'data/ml_rootwords.txt'))
self.dictionary = self.dictionary_file.readlines()
self.dictionary_file.close()
try:
self.dictionary = marisa_trie.Trie([x.strip().decode('utf-8')
for x in self.dictionary])
except:
self.dictionary = marisa_trie.Trie(
[x.strip() for x in self.dictionary])
self.stemmer = Stemmer()
self.inflector = inflector.Inflector(lang='ml')
self.soundex = Soundex()
self.syllabalizer = Syllabalizer()
self.ngrammer = Ngram()
def check(self, word):
"""
Returns if a word is spelled correctly or not.
"""
root_word = self.stemmer.stem(word)[word]['stem']
if root_word in self.dictionary:
return True
else:
return False
def get_best_intermediate(self, word, input_word,
intermediate_words, original_tag_list):
"""
Return the best intermediate form from those generated during stemming.
Best intermediate term is the one for which maximum similarity is
found. It is used to handle incorrect words getting unnecessarily
stemmed as they are not present in dictionary.
"""
lev = []
sound = []
jac = []
weight = []
word_tags_map = {}
selected_word = input_word
highest_weight = 0
for intr_counter in range(len(intermediate_words)):
intermediate_word = intermediate_words[intr_counter]
lev_tmp, sound_tmp, jac_tmp, weight_tmp = self.compare(
intermediate_word, word)
lev.append(lev_tmp)
sound.append(sound_tmp)
jac.append(jac_tmp)
weight.append(weight_tmp)
word_tags_map[intermediate_word] = original_tag_list[:intr_counter]
if len(weight) > 0:
highest_weight = max(weight)
position = weight.index(highest_weight)
selected_word = intermediate_words[position]
lev = lev[position]
return word_tags_map, highest_weight, selected_word
def get_unique(self, list_of_items):
result = []
for item in list_of_items:
if item not in result:
result.append(item)
return result
def suggest(self, input_word, n=5):
"""
Returns n suggestions that is similar to word.
"""
stemmer_result = self.stemmer.stem(input_word)[input_word]
input_word = stemmer_result['stem']
tag_list = stemmer_result['inflection']
first_char = input_word[0]
if first_char == _characters[0]:
prev_char = first_char
else:
prev_char_pos = _characters.index(first_char) - 1
prev_char = _characters[prev_char_pos]
if first_char == _characters[-1]:
next_char = first_char
else:
next_char_pos = _characters.index(first_char) + 1
next_char = _characters[next_char_pos]
possible_words = self.dictionary.keys(first_char) +\
self.dictionary.keys(next_char) +\
self.dictionary.keys(prev_char)
final = []
intermediate_words = []
original_tag_list = tag_list
intermediate_words.append(input_word)
for tag_counter in range(len(tag_list)):
new_word = self.inflector.inflect(
input_word, tag_list[-tag_counter - 1:])
intermediate_words.insert(0, new_word)
for word in possible_words:
lev, sound, jac, weight1 = self.compare(input_word, word)
word_tags_map, highest_weight, selected_word =\
self.get_best_intermediate(
word, input_word, intermediate_words, original_tag_list)
tag_list = original_tag_list
if highest_weight >= weight1 and selected_word != input_word:
tag_list = word_tags_map[selected_word]
weight = max(weight1, highest_weight)
suggestion_item = Malayalam.Suggestion(
word, sound, lev, jac, weight, tag_list)
if weight > 50:
final.append(suggestion_item)
sorted_list = sorted(final, key=attrgetter('weight'), reverse=True)[:n]
final_list = []
for item in sorted_list:
word = item.word
tag_list = item.tag_list
try:
inflected_form = self.inflector.inflect(word, tag_list)
final_list.append(inflected_form)
except:
final_list.append(word)
continue
return self.get_unique(final_list)
def levenshtein_distance(self, tokens1, tokens2):
"""
Takes two lists containing tokens of one word each and returns the
levenshtein distance between them.
"""
if len(tokens1) < len(tokens2):
return self.levenshtein_distance(tokens2, tokens1)
if len(tokens2) == 0:
return len(tokens1)
previous_row = range(len(tokens2) + 1)
for i, c1 in enumerate(tokens1):
current_row = [i + 1]
for j, c2 in enumerate(tokens2):
# j+1 instead of j since previous_row and current_row are one
# character longer
insertions = previous_row[j + 1] + 1
deletions = current_row[j] + 1 # than tokens2
substitutions = previous_row[j] + (c1 != c2)
current_row.append(min(insertions, deletions, substitutions))
previous_row = current_row
return previous_row[-1]
def compare(self, word1, word2):
"""
Returns the similarity measure between two words.
"""
soundex_comparison = self.soundex.compare(word1, word2)
tokens1 = self.syllabalizer.syllabify_ml(word1)
tokens2 = self.syllabalizer.syllabify_ml(word2)
levenshtein_distance = self.levenshtein_distance(tokens1, tokens2)
ngram1 = self.ngrammer.letterNgram(word1, 1)
ngram2 = self.ngrammer.letterNgram(word2, 1)
total = ngram1 + ngram2
union = []
for counter in range(len(total)):
item = total[counter]
if item not in union:
union.append(item)
final = [x for x in ngram1 if x in ngram2] +\
[x for x in ngram2 if x in ngram1]
intersection = []
for counter in range(len(final)):
item = final[counter]
if item not in intersection:
intersection.append(item)
jaccards = float(len(intersection)) / float(len(union))
if soundex_comparison == 1 or soundex_comparison == 0:
weight = 100
elif levenshtein_distance <= 2 and jaccards > 0.5:
weight = 75 + (1.5 * jaccards)
elif levenshtein_distance < 5 and jaccards > 0.5:
weight = 65 + (3 * jaccards)
else:
weight = 0
return levenshtein_distance, soundex_comparison, jaccards, weight
def check_and_generate(self, word):
"""
Receives a word as input, checks if it is a valid word and returns
the suggestions if it is not.
Returns 0 along with suggestions if an incorrect word.
Returns 1 along with blank list of suggestions if word in dictionary.
Returns 2 along with blank list of suggestions if word is unique.
"""
status = self.check(word)
if status:
return {'status': 1, 'suggestions': []}
else:
suggestions = self.suggest(word)
if suggestions:
return {'status': 0, 'suggestions': suggestions}
else:
# If there were no suggestions, it means the word was not
# similar to any of the existing root words. So, that was not a
# mistake, but an intended insertion. Hence, it is deemed as a
# valid word
return {'status': 2, 'suggestions': []}
phonemes2 = np.array([ipa[c] for c in pad2])
features = {'word 1 encoding': phonemes1, 'word 2 encoding': phonemes2}
return features
#%% Open some things
if __name__ == '__main__':
TRAIN_PATH = '../data/cognet_train.csv'
TEST_PATH = '../data/cognet_test.csv'
DEV_PATH = '../data/cognet_dev.csv'
DATA_PATH = '../data/extracted_features.npy'
SUPPORTED_LANGS_PATH = '../data/cognet_supported_langs.tsv'
IPA_ENCODING_PATH = '../data/ipa_encodings.pickle'
v = DictVectorizer(sparse=False)
soundex = Soundex()
epitran_dict = create_epitran_dict()
with open(IPA_ENCODING_PATH, 'rb') as f:
ipa = pickle.load(f)
ipa = defaultdict(lambda: np.array([0.] * 24), ipa)
#%% FEATURE EXTRACTION
if __name__ == '__main__':
print('Reading training data...')
train_data = pd.read_csv(TRAIN_PATH)
print('Extracting features...')
x_train = v.fit_transform([
extract_features(str(lang1), str(word1), str(lang2), str(word2))\
for lang1, word1, lang2, word2 in\
zip(train_data['lang 1'], train_data['translit 1'], train_data['lang 2'], train_data['translit 2'])
])
# -*- coding: utf-8 -*- """ Created on Sun Jun 18 00:04:21 2017 @author: Milton """ from foneticaBR import foneticaBR from buscabr import buscaBR from metaphoneBR import metaphoneBR # Soundex - Instalar no Python as bibliotecas soundex e silpa_common from soundex import Soundex chaveRoberto = foneticaBR() chavebr = buscaBR() chavemeta = metaphoneBR() chavesoundex = Soundex() texto = 'JOSSEPH' print(chaveRoberto.chavefonetica(texto)) print(chavebr.chaveBR(texto, False)) print(chavebr.chaveBR(texto, True)) print(chavemeta.chaveMetaphoneBR(texto)) print(chavesoundex.soundex(texto))
def sndx(s):
return Soundex().soundex(stemmer(s, args.stem))
from soundex import Soundex
from collections import defaultdict
from wordfreq import word_frequency
soundex = Soundex().soundex
sound_words = defaultdict(set)
with open('eff_short_wordlist_1.txt','r') as fh:
for line in fh:
word = line.split()[1]
sound = soundex(word)
if len(sound) > 1: # and sound not in ('i245', 't651'):
sound_words[sound].add(word)
for word_set in sound_words.values():
if len(word_set) > 1:
word_list = [ (word_frequency(word, 'en'), word) for word in word_set ]
word_list.sort()
print(word_list[-1][-1])
else:
print(list(word_set)[0])
'ben': SCHEMES[BENGALI],
'guj': SCHEMES[GUJARATI],
'hin': SCHEMES[DEVANAGARI],
'kan': SCHEMES[KANNADA],
'mal': SCHEMES[MALAYALAM],
'mar': SCHEMES[DEVANAGARI],
"ori": SCHEMES[ORIYA],
'pun': SCHEMES[GURMUKHI],
'tam': SCHEMES[TAMIL],
'tel': SCHEMES[TELUGU],
'urd': SCHEMES[DEVANAGARI]
}
if lang in scheme_dict:
src_scheme = scheme_dict[lang]
scheme_map = SchemeMap(src_scheme, SCHEMES[HK])
instance = Soundex()
findings = {}
for i, line in enumerate(inp_bible):
curr_line_id = 23146 + i
if line == "" or line == "\n":
continue
for index, name in enumerate(names_reference):
lids = ast.literal_eval(name[lid_col])
for col in col_references:
romans = name[col]
if romans == "":
continue
print(".", end="")
