class Zemberek:
def __init__(self,*args):
if len(args)>0:
JAVAPATH = args[0]
else:
JAVAPATH = getDefaultJVMPath()
print(f'JAVA PATH: {JAVAPATH}')
self.MAINFOLDER = os.path.dirname(os.path.realpath(__file__))
self.ZEMBEREK_PATH = os.path.join(self.MAINFOLDER,"zemberek-full.jar")
startJVM(JAVAPATH,'-ea',f'-Djava.class.path={self.ZEMBEREK_PATH}',convertStrings=False)
self.verbose = False
#
self.TurkishMorphology = JClass('zemberek.morphology.TurkishMorphology')
self.DictionaryItem = JClass('zemberek.morphology.lexicon.DictionaryItem')
self.RootAttribute = JClass('zemberek.core.turkish.RootAttribute')
self.PrimaryPos = JClass('zemberek.core.turkish.PrimaryPos')
self.SecondaryPos = JClass('zemberek.core.turkish.SecondaryPos')
self.WordAnalysis = JClass('zemberek.morphology.analysis.WordAnalysis')
self.AnalysisFormatters = JClass('zemberek.morphology.analysis.AnalysisFormatters')
self.RootLexicon = JClass('zemberek.morphology.lexicon.RootLexicon')
self.InformalAnalysisConverter = JClass('zemberek.morphology.analysis.InformalAnalysisConverter')
self.TurkishSentenceExtractor = JClass('zemberek.tokenization.TurkishSentenceExtractor')
self.TurkishTokenizer = JClass('zemberek.tokenization.TurkishTokenizer')
self.Token = JClass('zemberek.tokenization.Token')
self.TurkishSpellChecker = JClass('zemberek.normalization.TurkishSpellChecker')
self.TurkishSentenceNormalizer = JClass('zemberek.normalization.TurkishSentenceNormalizer')
self.PrimaryPos = JClass('zemberek.core.turkish.PrimaryPos')
self.SecondaryPos = JClass('zemberek.core.turkish.SecondaryPos')
# Derived ones
self.DefaultMorphology = self.TurkishMorphology.createWithDefaults() # Default Morphology..
self.InformalMorphology = self.TurkishMorphology.builder().setLexicon(self.RootLexicon.getDefault()).ignoreDiacriticsInAnalysis().useInformalAnalysis().build()
stopwordfile = os.path.join(self.MAINFOLDER,"turkish_stopwords.txt")
with open(stopwordfile, 'r',encoding = 'utf-8') as f :
StopWordsRead = f.readlines()
self.StopWords = list(map(removeNewLine,StopWordsRead))
f.close()
def __del__(self):
shutdownJVM()
def isStopWord(self,word) :
'''Is the given word is a stop word?'''
isstop = (word in self.StopWords) or word in '!"#$%&\'()*+,-./:;<=>?@[\\]^_`{|}~'
return isstop
def addNoun(self,noun_dict):
'''This function allows to add a new noun to dictionary.'''
dct = list(map(JString,noun_dict)) + [self.PrimaryPos.Noun, self.SecondaryPos.ProperNoun]
new_items = self.DictionaryItem(*dct)
self.DefaultMorphology.invalidateCache()
self.DefaultMorphology.getMorphotactics().getStemTransitions().addDictionaryItem(new_items)
def addVerb(self,noun_dict):
'''This function allows to add a new verb to dictionary.'''
dct = list(map(JString,noun_dict)) + [self.PrimaryPos.Verb, self.SecondaryPos.None_]
new_items = self.DictionaryItem(*dct)
self.DefaultMorphology.invalidateCache()
self.DefaultMorphology.getMorphotactics().getStemTransitions().addDictionaryItem(new_items)
def analyze(self,word):
'''This function analyzes the given word.'''
results = self.DefaultMorphology.analyze(JString(word))
if self.verbose:
for result in results :
print(
f'\nLexical and Surface: {str(result.formatLong())}'
f'\nOnly Lexical: {str(result.formatLexical())}'
'\nOflazer Style:'
f'{str(self.AnalysisFormatters.OFLAZER_STYLE.format(result))}'
)
return results, str(results)
def stemLemma(self,word):
'''This function stems and lemmatizes the given word.'''
results = self.DefaultMorphology.analyze(JString(word))
stems = []
lemmas = []
for result in results :
stems.append([str(stem) for stem in result.getStems()])
lemmas.append([str(stem) for stem in result.getLemmas()])
if self.verbose:
print(
f'{str(result.formatLong())}',
f'\n\tStems =', ' '.join(stems[-1]),
f'\n\tLemmas =', ' '.join(lemmas[-1]) )
return stems, lemmas
def sentenceDisambugation(self,sentence):
'''This function analyzes given a sentence.'''
stems = []
lemmas = []
if bool(sentence):
analysis = self.DefaultMorphology.analyzeSentence(sentence)
results = self.DefaultMorphology.disambiguate(sentence, analysis).bestAnalysis()
else:
return stems,lemmas
for i, result in enumerate(results, start=1) :
stems.append([str(stem) for stem in result.getStems()])
lemmas.append([str(stem) for stem in result.getLemmas()])
if self.verbose:
print(
f'\nAnalysis {i}: {str(result.formatLong())}'
f'\n\tStems =', ' '.join(stems[-1]),
f'\n\tLemmas =', ' '.join(lemmas[-1]) )
return stems, lemmas
def informalWordAnalysis(self,sentence):
'''This function analyzees informal words'''
analyses = self.InformalMorphology.analyzeAndDisambiguate(sentence).bestAnalysis() # : java.util.ArrayList
print('\nAnalysis:\n')
for analysis in analyses :
print(f'{str(analysis.surfaceForm())}-{analysis}')
print('\nConverting formal surface form:\n')
converter = self.InformalAnalysisConverter(self.InformalMorphology.getWordGenerator())
if self.verbose:
for analysis in analyses :
print(str(converter.convert(analysis.surfaceForm(), analysis)))
return list(map(str, analyses))
def findPOS(self,sentence):
'''This function finds parts of speech in a given sentence.'''
analysis = self.DefaultMorphology.analyzeAndDisambiguate(sentence).bestAnalysis()
pos = []
for i, analysis in enumerate(analysis, start=1) :
if self.verbose:
print(
f'\nAnalysis {i}: {analysis}',
f'\nPrimary POS {i}: {analysis.getPos()}'
f'\nPrimary POS (Short Form) {i}: {analysis.getPos().shortForm}')
pos.append(
f'{str(analysis.getLemmas()[0])}'
f'-{analysis.getPos().shortForm}'
)
if self.verbose:
print(f'\nFull sentence with POS tags: {" ".join(pos)}')
return pos
def sentenceBoundary(self,paragraph):
'''This function detects bounds of a sentence.'''
sentences = self.TurkishSentenceExtractor.DEFAULT.fromParagraph(paragraph)
if self.verbose:
for i, sent in enumerate(sentences) :
print(f'Sentence {i + 1}: {sent}')
return list(map(str,sentences)) # Return lists
def sentenceTokenization(self,sentence):
'''This function tokenizes a simple sentence.'''
token_iterator = self.TurkishTokenizer.DEFAULT.tokenizeToStrings(JString(sentence))
if self.verbose:
print('\nToken Iterator Example:\n')
for i, token in enumerate(token_iterator) :
print(f'Token {i} = {token}')
return list(map(str,token_iterator))
# Normalization functions.
def correctDocument(self,document):
'''This function corrects misspelled words in a document.'''
spell_checker = self.TurkishSpellChecker(self.DefaultMorphology)
tokens = self.TurkishTokenizer.ALL.tokenize(JString(document))
corrected_tokens = []
for token in tokens :
text = token.content
if (
token.type != self.Token.Type.NewLine
and token.type != self.Token.Type.SpaceTab
and token.type != self.Token.Type.Punctuation
and token.type != self.Token.Type.RomanNumeral
and token.type != self.Token.Type.UnknownWord
and token.type != self.Token.Type.Unknown
and not spell_checker.check(text)
) :
suggestions = list(spell_checker.suggestForWord(token.content))
if suggestions :
suggestion: str = str(suggestions[0])
print(f'Correction: {token.content} -> {suggestion}.')
corrected_tokens.append(suggestion)
continue
corrected_tokens.append(str(token.content))
correctedDoc = ' '.join(corrected_tokens)
if self.verbose:
print('\nCorrected Document:\n', correctedDoc)
return correctedDoc
def normalizeDocument(self,document):
'''This function normalizes a given document.'''
Paths: JClass = JClass('java.nio.file.Paths')
path1 = Paths.get(os.path.join('.', 'req_data'))
path2 = Paths.get(os.path.join('.', 'req_data', 'lm.2gram.slm'))
normalizer = self.TurkishSentenceNormalizer(self.TurkishMorphology.createWithDefaults(),path1,path2)
normalizedDoc = normalizer.normalize(JString(document))
if self.verbose:
print(f'\nNoisy : {document}')
print(f'\nNormalized : {normalizedDoc}' )
return str(normalizedDoc)
def NER(self,sentence):
'''This function performs Named Entity Recognition for a sentence.'''
# Use pre-trained model.
pass
def analyze2(self,word,add_swt=True):
'''
Extract possible subword combinations from a word.
add_swt: Add additive sub-words starting with ##
'''
analysis_all, _ = self.analyze(word)
#if len(analysis_all) == 0:
# return [[word]]
allanalysis = []
for analysis_single in analysis_all:
stranal = str(analysis_single)
strng = re.sub('\[\S+:\S+\] ','',stranal)
parts = re.split('[|\+]', strng)
isNotFirst = False
partanalysis = []
for part in parts:
if ':' in part:
root = re.split(':',part)[0]
if isNotFirst and not add_swt:
break
elif isNotFirst:
root = '##' + root
else:
isNotFirst = True # Remaining parts are not first..
partanalysis.append(root)
allanalysis.append(partanalysis)
return allanalysis
def uniqueSubWords(self,word):
'''This function returns all possible subwords of given word as a SET.'''
analyzed = self.analyze2(word)
subwords = [subword for analyze_x in analyzed for subword in analyze_x]
return set(subwords)
def normalizeMultipleSentences(self, paragraph):
# Remove newline chaacters from the document
paraflat = removeNewLine(paragraph)
# Split sentences
para_divd = self.sentenceBoundary(paraflat)
# Normalize
n_fcn = lambda x: normalizeText(x,False,False,True).split()
norm_txt = list(map(n_fcn,para_divd))
return norm_txt
# This function may be removed..
def parseWords(self, paragraph, add_swt=True, rmnewline=True, lwcase=True, expunc=True):
'''
Returns dictionary of word counts.
add_swt: Add additive sub-words starting with ##
rmnewline: Remove new lines.
lwcase: Lowercase text.
expunc: Expand punctuation.
'''
para = normalizeText(paragraph,rmnewline,lwcase,expunc)
words = para.split()
count_dict = {}
norm_txt = ''
for word in words:
# print(f'size: {len(self.analyze2(word))} \n')
a2 = self.analyze2(word,add_swt)
if len(a2) == 0: # If word cannot be analyzed, pass as it is.
# print(f'not parsed word: {word}')
txt = word
analysis = [txt]
else:
# get longest first word as analysis...
a_lens = list(map(lambda x: len(x[0]),a2)) # analysis lengths.
idx = a_lens.index(max(a_lens))
analysis = a2[idx] # get first analysis.
if not add_swt and self.isStopWord(analysis[0]): # If word is a stop-word, and additive words are disregard.
continue
txt = ' '.join(analysis)
for a in analysis:
if a not in count_dict.keys():
count_dict[a]=1
else:
count_dict[a] += 1
norm_txt = norm_txt + ' ' + txt
return norm_txt, count_dict
from sklearn.neural_network import MLPClassifier
from sklearn.utils.extmath import density
from sklearn import metrics
from sklearn.linear_model import LogisticRegression
from sklearn.model_selection import GridSearchCV
# import lemmatizer
import pickle
from typing import List
from jpype import JClass, JString, getDefaultJVMPath, shutdownJVM, startJVM, java
from unidecode import unidecode
ZEMBEREK_PATH = r'C:\Users\golive\Desktop\nlp_yazi\zemberek-all-0.11.1.jar'
startJVM(getDefaultJVMPath(), '-ea', '-Djava.class.path=%s' % (ZEMBEREK_PATH))
turkish_morphology = JClass('zemberek.morphology.TurkishMorphology')
morphology = turkish_morphology.createWithDefaults()
# import os
# os.environ['LDFLAGS'] = '-framework CoreFoundation -framework SystemConfiguration'
# !pip3 install spacy
print(sys.path)
nlp = spacy.load('en_core_web_lg')
# print('Running')
try:
with open('EsAnlamlilar.csv', 'r', encoding="utf-8") as f:
reader = csv.reader(f, delimiter=',')
mydict = {key: row for key, *row in reader}
except IOError:
import os
import re
import nltk
nltk.download("punkt")
import xml.etree.ElementTree as ET
from typing import List
from jpype import JClass, JString, getDefaultJVMPath, shutdownJVM, startJVM, java
ZEMBEREK_PATH = './zemberek-full.jar'
startJVM(getDefaultJVMPath(), '-ea', '-Djava.class.path=%s' % (ZEMBEREK_PATH))
TurkishMorphology = JClass('zemberek.morphology.TurkishMorphology')
morphology = TurkishMorphology.createWithDefaults()
path = './Makaleler'
for filename in os.listdir(path):
fullname = os.path.join(path, filename)
tree = ET.parse(fullname)
root = tree.getroot()
for ozetce in root:
ozetce = ozetce.find('Özetçe').text
ozetce = str(ozetce)
kelimeler = nltk.word_tokenize(ozetce)
yeni_kelimeler= [kelimeler for kelimeler in kelimeler if kelimeler.isalnum()]
pos: List[str] = []
for kelime in yeni_kelimeler:
analysis: java.util.ArrayList = (
morphology.analyzeAndDisambiguate(kelime).bestAnalysis()
)
