fast aligner

fast-aligner/Makefile

@@ -5,16 +5,23 @@ SEPARATOR=@\#@
 
 DICTIONARY_WEIGHT=3
 
-all: corpora/$(CORPUS_NAME)/falign_corpus.txt
+all: corpora/$(CORPUS_NAME)/alignments.txt corpora/$(CORPUS_NAME)/src_clean.txt corpora/$(CORPUS_NAME)/src_clean.lem corpora/$(CORPUS_NAME)/trg_clean.txt
 
 
 clean:
 	rm -f corpora/$(CORPUS_NAME)/*.lem
 	rm -f corpora/$(CORPUS_NAME)/*.dict
 	rm -f corpora/$(CORPUS_NAME)/src_clean.txt
+	rm -f corpora/$(CORPUS_NAME)/src_clean.lem
 	rm -f corpora/$(CORPUS_NAME)/trg_clean.txt
 	rm -f corpora/$(CORPUS_NAME)/ids_clean.txt
 	rm -f corpora/$(CORPUS_NAME)/falign_corpus.txt
+	rm -f corpora/$(CORPUS_NAME)/falign_result.txt
+	rm -f corpora/$(CORPUS_NAME)/alignments.txt
+
+
+corpora/$(CORPUS_NAME)/alignments.txt: corpora/$(CORPUS_NAME)/falign_result.txt corpora/$(CORPUS_NAME)/src_clean.lem
+	./get_alignments.py corpora/$(CORPUS_NAME)/falign_result.txt corpora/$(CORPUS_NAME)/src_clean.lem > $@
 
 corpora/$(CORPUS_NAME)/src.dict:
 	./collect_dict.py $(SRC_LANG) $(TRG_LANG) $(DICTIONARY_WEIGHT) > $@
@@ -30,16 +37,9 @@ corpora/$(CORPUS_NAME)/trg.lem: corpora/$(CORPUS_NAME)/trg.txt
 	/usr/local/bin/concordia-sentence-tokenizer -c ../concordia.cfg < $< | ./sentence_lemmatizer.py $(TRG_LANG) > $@
 
 
+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/$(CORPUS_NAME)/falign_corpus.txt: corpora/$(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
+	./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
 
 
-corpora/$(CORPUS_NAME)/src_clean.txt corpora/$(CORPUS_NAME)/trg_clean.txt corpora/$(CORPUS_NAME)/ids_clean.txt corpora/$(CORPUS_NAME)/falign_corpus.txt: 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)/falign_result.txt: corpora/$(CORPUS_NAME)/falign_corpus.txt 
-	./clean_corpus.py corpora/$(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/$(CORPUS_NAME)/src_clean.txt corpora/$(CORPUS_NAME)/trg_clean.txt corpora/$(CORPUS_NAME)/ids_clean.txt corpora/$(CORPUS_NAME)/falign_corpus.txt
+	fast_align -i $< -d -o -v > $@
-
-corpora/$(CORPUS_NAME)/pasted.txt: corpora/$(CORPUS_NAME)/src_clean.txt corpora/$(CORPUS_NAME)/trg_clean.txt
-	./paste.py corpora/$(CORPUS_NAME)/src_clean.txt corpora/$(CORPUS_NAME)/trg_clean.txt $(SEPARATOR)> $@
-
-corpora/$(CORPUS_NAME)/pasted_deduplicated.txt: corpora/$(CORPUS_NAME)/pasted.txt
-	sort -k 1.13 $< | uniq -s 12 | sort > $@
-
-corpora/$(CORPUS_NAME)/aligned_final.txt corpora/$(CORPUS_NAME)/src_final.txt corpora/$(CORPUS_NAME)/trg_final.txt: corpora/$(CORPUS_NAME)/pasted_deduplicated.txt corpora/$(CORPUS_NAME)/aligned.txt
-	./extract.py $< corpora/$(CORPUS_NAME)/aligned.txt corpora/$(CORPUS_NAME)/aligned_final.txt corpora/$(CORPUS_NAME)/src_final.txt corpora/$(CORPUS_NAME)/trg_final.txt $(SEPARATOR)

fast-aligner/collect_dict.py

@@ -7,9 +7,11 @@ src_lang = sys.argv[1]
 trg_lang = sys.argv[2]
 weight = int(sys.argv[3])
 
-for dname in os.listdir('dictionaries'):
+dictionaries_path = '../dictionaries'
-    src_path = 'dictionaries/%s/%s.bz2' % (dname, src_lang)
+
-    trg_path = 'dictionaries/%s/%s.bz2' % (dname, trg_lang)
+for dname in os.listdir(dictionaries_path):
+    src_path = '%s/%s/%s.bz2' % (dictionaries_path, dname, src_lang)
+    trg_path = '%s/%s/%s.bz2' % (dictionaries_path, dname, trg_lang)
     if os.path.isfile(src_path) and os.path.isfile(trg_path):
         with bz2.open(src_path, 'rt') as src_dict_file:
             for line in src_dict_file:

fast-aligner/get_alignments.py

@@ -0,0 +1,25 @@
+#!/usr/bin/python3
+# -*- coding: utf-8 -*-
+
+import sys
+
+
+with open(sys.argv[1]) as falign_result, open(sys.argv[2]) as src_clean_lem:
+    for line in src_clean_lem:
+        src_line_lem = line.strip()
+        falign_alignment_pairs = falign_result.readline().rstrip().split()
+        falign_map = {}
+        for pair in falign_alignment_pairs:
+            numbers = pair.split('-')
+            s = int(numbers[0])
+            t = int(numbers[1])
+            if not s in falign_map:
+                falign_map[s] = []
+            falign_map[s].append(t)
+        res = []
+        for i in range(len(src_line_lem.split())):
+            if i in falign_map:
+                res.append(falign_map[i])
+            else:
+                res.append([])
+        print(res)

fast-aligner/prepare_corpus.py

@@ -6,10 +6,12 @@ import sys
 max_tokens = 100
 max_ratio = 4.0
 
-#./clean_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_clean.txt corpora/$(CORPUS_NAME)/trg_clean.txt corpora/$(CORPUS_NAME)/ids_clean.txt corpora/$(CORPUS_NAME)/falign_corpus.txt
+#./prepare_corpus.py corpora/$(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
 
 
-with open(sys.argv[1]) as src, open(sys.argv[2]) as trg, open(sys.argv[3]) as ids, open(sys.argv[4]) as src_lem, open(sys.argv[5]) as trg_lem, open(sys.argv[6], 'w') as src_clean, open(sys.argv[7], 'w') as trg_clean, open(sys.argv[8], 'w') as ids_clean, open(sys.argv[9], 'w') as falign_corpus:
+
+
+with open(sys.argv[1]) as src, open(sys.argv[2]) as trg, open(sys.argv[3]) as ids, open(sys.argv[4]) as src_lem, open(sys.argv[5]) as trg_lem, open(sys.argv[6]) as src_dict, open(sys.argv[7]) as trg_dict, open(sys.argv[8], 'w') as src_clean, open(sys.argv[9], 'w') as src_clean_lem, open(sys.argv[10], 'w') as trg_clean, open(sys.argv[11], 'w') as ids_clean, open(sys.argv[12], 'w') as falign_corpus:
     for line in src:
         src_line_orig = line.strip()
         trg_line_orig = trg.readline().strip()
@@ -22,6 +24,11 @@ with open(sys.argv[1]) as src, open(sys.argv[2]) as trg, open(sys.argv[3]) as id
             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):
                 src_clean.write(src_line_orig+"\n")
+                src_clean_lem.write(src_line_lem+"\n")
                 trg_clean.write(trg_line_orig+"\n")
                 ids_clean.write(id_orig+"\n")
-                falign_corpus.write("%s|||%s\n" % (src_line_lem, trg_line_lem))
+                falign_corpus.write("%s ||| %s\n" % (src_line_lem, trg_line_lem))
+    for line in src_dict:
+        src_word = line.rstrip()
+        trg_word = trg_dict.readline().rstrip()
+        falign_corpus.write("%s ||| %s\n" % (src_word, trg_word))