better bad words filtering

fast-aligner/bad-words/parts.en

@@ -0,0 +1,3 @@
+fuck
+shit
+nigga

fast-aligner/bad-words/parts.pl

@@ -0,0 +1,4 @@
+pierdol
+pierdal
+kurw
+ścierw

fast-aligner/bad-words/whole.en

@@ -1,9 +1,14 @@
 arse
 ass
+asshole
+assholes
 ballsack
 bastard
 biatch
 bitch
+bitchcraft
+bitchen
+bitchy
 blowjob
 bollock
 bollok
@@ -17,6 +22,12 @@ crap
 cunt
 damn
 dick
+dicker
+dickerin
+dickhead
+dickheads
+dickweed
+dickweeds
 dildo
 dyke
 fag
@@ -30,10 +41,13 @@ fuck
 fudgepacker
 Goddamn
 hell
+Horsedick
 jizz
 knobend
 muff
+needledick
 nigga
+niggas
 nigger
 penis
 piss
@@ -44,6 +58,8 @@ scrotum
 sh1t
 shit
 slut
+sluttish
+slutty
 spunk
 toss
 tosser
@@ -51,4 +67,7 @@ turd
 twat
 vagina
 wank
+wanker
 whore
+whorehouse
+whoremonger

fast-aligner/prepare_corpus.py

@@ -4,15 +4,24 @@
 import sys, re, os
 
 
-def readProfanityPattern(lang):
+def readProfanityPattern(lang, whole):
     result = None
-    file_path = 'bad-words/%s.txt' % lang
+    if whole:
+        file_path = 'bad-words/whole.'+ lang
+    else:
+        file_path = 'bad-words/parts.'+ lang        
     if os.path.isfile(file_path):
-        raw_pattern = r'\b('
+        raw_pattern = ''
+        if not whole:
+            raw_pattern = r'\b'
+        raw_pattern += '('
         with open(file_path) as profanity_file:
             for line in profanity_file:
                 raw_pattern += line.rstrip() + "|"
-        raw_pattern = raw_pattern[:-1] + r')\b'
+        raw_pattern = raw_pattern[:-1] + ')'
+        if not whole:
+            raw_pattern += r'\b'
+
         result = re.compile(raw_pattern)
     return result
 
@@ -32,9 +41,13 @@ max_ratio = 4.0
 
 #./prepare_corpus.py corpora/$(CORPUS_NAME)/src.txt corpora/$(CORPUS_NAME)/trg.txt corpora/$(CORPUS_NAME)/ids.txt corpora/$(CORPUS_NAME)/src.lem corpora/$(CORPUS_NAME)/trg.lem corpora/$(CORPUS_NAME)/src.dict corpora/$(CORPUS_NAME)/trg.dict corpora/$(CORPUS_NAME)/src_clean.txt corpora/$(CORPUS_NAME)/src_clean.lem corpora/$(CORPUS_NAME)/trg_clean.txt corpora/$(CORPUS_NAME)/ids_clean.txt corpora/$(CORPUS_NAME)/falign_corpus.txt $SRC_LANG $TRG_LANG
 
-src_profanity_pattern = readProfanityPattern(sys.argv[13])
-trg_profanity_pattern = readProfanityPattern(sys.argv[14])
+src_lang = sys.argv[13]
+trg_lang = sys.argv[14]
 
+src_profanity_whole_pattern = readProfanityPattern(src_lang, True)
+src_profanity_parts_pattern = readProfanityPattern(src_lang, False)
+trg_profanity_whole_pattern = readProfanityPattern(trg_lang, True)
+trg_profanity_parts_pattern = readProfanityPattern(trg_lang, False)
 
 
 
@@ -50,7 +63,12 @@ with open(sys.argv[1]) as src, open(sys.argv[2]) as trg, open(sys.argv[3]) as id
         trg_token_count = len(trg_line_lem.split())
         if (src_token_count > 0 and trg_token_count > 0 and src_token_count <= max_tokens and trg_token_count <= max_tokens):
             ratio = float(src_token_count/trg_token_count) if src_token_count > trg_token_count else float(trg_token_count/src_token_count)
-            if (ratio <= max_ratio and (not containsProfanity(src_profanity_pattern, src_line_lem)) and (not containsProfanity(trg_profanity_pattern, trg_line_lem))):
+            if (ratio <= max_ratio 
+                   and (not containsProfanity(src_profanity_whole_pattern, src_line_lem))
+                   and (not containsProfanity(src_profanity_parts_pattern, src_line_lem))
+                   and (not containsProfanity(trg_profanity_whole_pattern, trg_line_lem))
+                   and (not containsProfanity(trg_profanity_parts_pattern, trg_line_lem))
+                ):
                 src_clean.write(src_line_orig+"\n")
                 src_clean_lem.write(src_line_lem+"\n")
                 trg_clean.write(trg_line_orig+"\n")