# -*- coding: utf-8 -*- # # Natural Language Toolkit: The ISRI Arabic Stemmer # # Copyright (C) 2001-2010 NLTK Proejct # Algorithm: Kazem Taghva, Rania Elkhoury, and Jeffrey Coombs (2005) # Author: Hosam Algasaier # URL: # For license information, see LICENSE.TXT """ISRI Arabic Stemmer The algorithm for this stemmer is described in: Taghva, K., Elkoury, R., and Coombs, J. 2005. Arabic Stemming without a root dictionary. Information Science Research Institute. University of Nevada, Las Vegas, USA. The Information Science Research Institute’s (ISRI) Arabic stemmer shares many features with the Khoja stemmer. However, the main difference is that ISRI stemmer does not use root dictionary. Also, if a root is not found, ISRI stemmer returned normalized form, rather than returning the original unmodified word. Additional adjustments were made to improve the algorithm: 1- Adding 60 stop words. 2- Adding the pattern (تفاعيل) to ISRI pattern set. 3- The step 2 in the original algorithm was normalizing all hamza. This step is discarded because it increases the word ambiguities and changes the original root. """ import re from api import * class ISRIStemmer(StemmerI): ''' ISRI Arabic stemmer based on algorithm: Arabic Stemming without a root dictionary. Information Science Research Institute. University of Nevada, Las Vegas, USA. A few minor modifications have been made to ISRI basic algorithm. See the source code of this module for more information. isri.stem(token) returns Arabic root for the given token. The ISRI Stemmer requires that all tokens have Unicode string types. If you use Python IDLE on Arabic Windows you have to decode text first using Arabic '1256' coding. ''' def __init__(self): self.stm = 'defult none' self.p3 = [u'\u0643\u0627\u0644', u'\u0628\u0627\u0644', u'\u0648\u0644\u0644', u'\u0648\u0627\u0644'] # length three prefixes self.p2 = [u'\u0627\u0644', u'\u0644\u0644'] # length two prefixes self.p1 = [u'\u0644', u'\u0628', u'\u0641', u'\u0633', u'\u0648', u'\u064a', u'\u062a', u'\u0646', u'\u0627'] # length one prefixes self.s3 = [u'\u062a\u0645\u0644', u'\u0647\u0645\u0644', u'\u062a\u0627\u0646', u'\u062a\u064a\u0646', u'\u0643\u0645\u0644'] # length three suffixes self.s2 = [u'\u0648\u0646', u'\u0627\u062a', u'\u0627\u0646', u'\u064a\u0646', u'\u062a\u0646', u'\u0643\u0645', u'\u0647\u0646', u'\u0646\u0627', u'\u064a\u0627', u'\u0647\u0627', u'\u062a\u0645', u'\u0643\u0646', u'\u0646\u064a', u'\u0648\u0627', u'\u0645\u0627', u'\u0647\u0645'] # length two suffixes self.s1 = [u'\u0629', u'\u0647', u'\u064a', u'\u0643', u'\u062a', u'\u0627', u'\u0646'] # length one suffixes self.pr4 = {0:[u'\u0645'], 1:[u'\u0627'], 2:[u'\u0627', u'\u0648', u'\u064A'], 3:[u'\u0629']} # groups of length four patterns self.pr53 = {0:[u'\u0627', u'\u062a'], 1:[u'\u0627', u'\u064a', u'\u0648'], 2:[u'\u0627', u'\u062a', u'\u0645'], 3:[u'\u0645', u'\u064a', u'\u062a'], 4:[u'\u0645', u'\u062a'], 5:[u'\u0627', u'\u0648'], 6:[u'\u0627', u'\u0645']} # Groups of length five patterns and length three roots self.re_short_vowels = re.compile(ur'[\u064B-\u0652]') self.re_hamza = re.compile(ur'[\u0621\u0624\u0626]') self.re_intial_hamza = re.compile(ur'^[\u0622\u0623\u0625]') self.stop_words = [u'\u064a\u0643\u0648\u0646', u'\u0648\u0644\u064a\u0633', u'\u0648\u0643\u0627\u0646', u'\u0643\u0630\u0644\u0643', u'\u0627\u0644\u062a\u064a', u'\u0648\u0628\u064a\u0646', u'\u0639\u0644\u064a\u0647\u0627', u'\u0645\u0633\u0627\u0621', u'\u0627\u0644\u0630\u064a', u'\u0648\u0643\u0627\u0646\u062a', u'\u0648\u0644\u0643\u0646', u'\u0648\u0627\u0644\u062a\u064a', u'\u062a\u0643\u0648\u0646', u'\u0627\u0644\u064a\u0648\u0645', u'\u0627\u0644\u0644\u0630\u064a\u0646', u'\u0639\u0644\u064a\u0647', u'\u0643\u0627\u0646\u062a', u'\u0644\u0630\u0644\u0643', u'\u0623\u0645\u0627\u0645', u'\u0647\u0646\u0627\u0643', u'\u0645\u0646\u0647\u0627', u'\u0645\u0627\u0632\u0627\u0644', u'\u0644\u0627\u0632\u0627\u0644', u'\u0644\u0627\u064a\u0632\u0627\u0644', u'\u0645\u0627\u064a\u0632\u0627\u0644', u'\u0627\u0635\u0628\u062d', u'\u0623\u0635\u0628\u062d', u'\u0623\u0645\u0633\u0649', u'\u0627\u0645\u0633\u0649', u'\u0623\u0636\u062d\u0649', u'\u0627\u0636\u062d\u0649', u'\u0645\u0627\u0628\u0631\u062d', u'\u0645\u0627\u0641\u062a\u0626', u'\u0645\u0627\u0627\u0646\u0641\u0643', u'\u0644\u0627\u0633\u064a\u0645\u0627', u'\u0648\u0644\u0627\u064a\u0632\u0627\u0644', u'\u0627\u0644\u062d\u0627\u0644\u064a', u'\u0627\u0644\u064a\u0647\u0627', u'\u0627\u0644\u0630\u064a\u0646', u'\u0641\u0627\u0646\u0647', u'\u0648\u0627\u0644\u0630\u064a', u'\u0648\u0647\u0630\u0627', u'\u0644\u0647\u0630\u0627', u'\u0641\u0643\u0627\u0646', u'\u0633\u062a\u0643\u0648\u0646', u'\u0627\u0644\u064a\u0647', u'\u064a\u0645\u0643\u0646', u'\u0628\u0647\u0630\u0627', u'\u0627\u0644\u0630\u0649'] def stem(self, token): """ Stemming a word token using the ISRI stemmer. """ self.stm = token self.norm(1) # remove diacritics which representing Arabic short vowels if self.stm in self.stop_words: return self.stm # exclude stop words from being processed self.pre32() # remove length three and length two prefixes in this order self.suf32() # remove length three and length two suffixes in this order self.waw() # remove connective ‘و’ if it precedes a word beginning with ‘و’ self.norm(2) # normalize initial hamza to bare alif if len(self.stm)<=3: return self.stm # return stem if less than or equal to three if len(self.stm)==4: # length 4 word self.pro_w4() return self.stm elif len(self.stm)==5: # length 5 word self.pro_w53() self.end_w5() return self.stm elif len(self.stm)==6: # length 6 word self.pro_w6() self.end_w6() return self.stm elif len(self.stm)==7: # length 7 word self.suf1() if len(self.stm)==7: self.pre1() if len(self.stm)==6: self.pro_w6() self.end_w6() return self.stm return self.stm # if word length >7 , then no stemming def norm(self, num): """ normalization: num=1 normalize diacritics num=2 normalize initial hamza num=3 both 1&2 """ self.k = num if self.k == 1: self.stm = self.re_short_vowels.sub('', self.stm) return self.stm elif self.k == 2: self.stm = self.re_intial_hamza.sub(ur'\u0627',self.stm) return self.stm elif self.k == 3: self.stm = self.re_short_vowels.sub('', self.stm) self.stm = self.re_intial_hamza.sub(ur'\u0627',self.stm) return self.stm def pre32(self): """remove length three and length two prefixes in this order""" if len(self.stm)>=6: for pre3 in self.p3: if self.stm.startswith(pre3): self.stm = self.stm[3:] return self.stm elif len(self.stm)>=5: for pre2 in self.p2: if self.stm.startswith(pre2): self.stm = self.stm[2:] return self.stm def suf32(self): """remove length three and length two suffixes in this order""" if len(self.stm)>=6: for suf3 in self.s3: if self.stm.endswith(suf3): self.stm = self.stm[:-3] return self.stm elif len(self.stm)>=5: for suf2 in self.s2: if self.stm.endswith(suf2): self.stm = self.stm[:-2] return self.stm def waw(self): """remove connective ‘و’ if it precedes a word beginning with ‘و’ """ if (len(self.stm)>=4)&(self.stm[:2] == u'\u0648\u0648'): self.stm = self.stm[1:] return self.stm def pro_w4(self): """process length four patterns and extract length three roots""" if self.stm[0] in self.pr4[0]: # مفعل self.stm = self.stm[1:] return self.stm elif self.stm[1] in self.pr4[1]: # فاعل self.stm = self.stm[0]+self.stm[2:] return self.stm elif self.stm[2] in self.pr4[2]: # فعال - فعول - فعيل self.stm = self.stm[:2]+self.stm[3] return self.stm elif self.stm[3] in self.pr4[3]: # فعلة self.stm = self.stm[:-1] return self.stm else: self.suf1() # do - normalize short sufix if len(self.stm)==4: self.pre1() # do - normalize short prefix return self.stm def pro_w53(self): """process length five patterns and extract length three roots""" if ((self.stm[2] in self.pr53[0]) & (self.stm[0] == u'\u0627')): # افتعل - افاعل self.stm = self.stm[1]+self.stm[3:] return self.stm elif ((self.stm[3] in self.pr53[1]) & (self.stm[0] == u'\u0645')): # مفعول - مفعال - مفعيل self.stm = self.stm[1:3]+self.stm[4] return self.stm elif ((self.stm[0] in self.pr53[2]) & (self.stm[4] == u'\u0629')): # مفعلة - تفعلة - افعلة self.stm = self.stm[1:4] return self.stm elif ((self.stm[0] in self.pr53[3]) & (self.stm[2] == u'\u062a')): # مفتعل - يفتعل - تفتعل self.stm = self.stm[1]+self.stm[3:] return self.stm elif ((self.stm[0] in self.pr53[4]) & (self.stm[2] == u'\u0627')): #مفاعل - تفاعل self.stm = self.stm[1]+self.stm[3:] return self.stm elif ((self.stm[2] in self.pr53[5]) & (self.stm[4] == u'\u0629')): # فعولة - فعالة self.stm = self.stm[:2]+self.stm[3] return self.stm elif ((self.stm[0] in self.pr53[6]) & (self.stm[1] == u'\u0646')): # انفعل - منفعل self.stm = self.stm[2:] return self.stm elif ((self.stm[3] == u'\u0627') & (self.stm[0] == u'\u0627')): # افعال self.stm = self.stm[1:3]+self.stm[4] return self.stm elif ((self.stm[4] == u'\u0646') & (self.stm[3] == u'\u0627')): # فعلان self.stm = self.stm[:3] return self.stm elif ((self.stm[3] == u'\u064a') & (self.stm[0] == u'\u062a')): # تفعيل self.stm = self.stm[1:3]+self.stm[4] return self.stm elif ((self.stm[3] == u'\u0648') & (self.stm[1] == u'\u0627')): # فاعول self.stm = self.stm[0]+self.stm[2]+self.stm[4] return self.stm elif ((self.stm[2] == u'\u0627') & (self.stm[1] == u'\u0648')): # فواعل self.stm = self.stm[0]+self.stm[3:] return self.stm elif ((self.stm[3] == u'\u0626') & (self.stm[2] == u'\u0627')): # فعائل self.stm = self.stm[:2]+self.stm[4] return self.stm elif ((self.stm[4] == u'\u0629') & (self.stm[1] == u'\u0627')): # فاعلة self.stm = self.stm[0]+self.stm[2:4] return self.stm elif ((self.stm[4] == u'\u064a') & (self.stm[2] == u'\u0627')): # فعالي self.stm = self.stm[:2]+self.stm[3] return self.stm else: self.suf1() # do - normalize short sufix if len(self.stm)==5: self.pre1() # do - normalize short prefix return self.stm def pro_w54(self): """process length five patterns and extract length four roots""" if (self.stm[0] in self.pr53[2]): #تفعلل - افعلل - مفعلل self.stm = self.stm[1:] return self.stm elif (self.stm[4] == u'\u0629'): # فعللة self.stm = self.stm[:4] return self.stm elif (self.stm[2] == u'\u0627'): # فعالل self.stm = self.stm[:2]+self.stm[3:] return self.stm def end_w5(self): """ending step (word of length five)""" if len(self.stm)==3: return self.stm elif len(self.stm)==4: self.pro_w4() return self.stm elif len(self.stm)==5: self.pro_w54() return self.stm def pro_w6(self): """process length six patterns and extract length three roots""" if ((self.stm.startswith(u'\u0627\u0633\u062a')) or (self.stm.startswith(u'\u0645\u0633\u062a'))): # مستفعل - استفعل self.stm= self.stm[3:] return self.stm elif (self.stm[0]== u'\u0645' and self.stm[3]== u'\u0627' and self.stm[5]== u'\u0629'): # مفعالة self.stm = self.stm[1:3]+self.stm[4] return self.stm elif (self.stm[0]== u'\u0627' and self.stm[2]== u'\u062a' and self.stm[4]== u'\u0627'): # افتعال self.stm = self.stm[1]+self.stm[3]+self.stm[5] return self.stm elif (self.stm[0]== u'\u0627' and self.stm[3]== u'\u0648' and self.stm[2]==self.stm[4]): # افعوعل self.stm = self.stm[1]+self.stm[4:] return self.stm elif (self.stm[0]== u'\u062a' and self.stm[2]== u'\u0627' and self.stm[4]== u'\u064a'): # تفاعيل new pattern self.stm = self.stm[1]+self.stm[3]+self.stm[5] return self.stm else: self.suf1() # do - normalize short sufix if len(self.stm)==6: self.pre1() # do - normalize short prefix return self.stm def pro_w64(self): """process length six patterns and extract length four roots""" if (self.stm[0] and self.stm[4])==u'\u0627': # افعلال self.stm=self.stm[1:4]+self.stm[5] return self.stm elif (self.stm.startswith(u'\u0645\u062a')): # متفعلل self.stm = self.stm[2:] return self.stm def end_w6(self): """ending step (word of length six)""" if len(self.stm)==3: return self.stm elif len(self.stm)==5: self.pro_w53() self.end_w5() return self.stm elif len (self.stm)==6: self.pro_w64() return self.stm def suf1(self): """normalize short sufix""" for sf1 in self.s1: if self.stm.endswith(sf1): self.stm = self.stm[:-1] return self.stm def pre1(self): """normalize short prefix""" for sp1 in self.p1: if self.stm.startswith(sp1): self.stm = self.stm[1:] return self.stm