concordia-server/fast-aligner/prepare_corpus.py

#!/usr/bin/python3
# -*- coding: utf-8 -*-

import sys, re, os


def readProfanityPattern(lang):
    result = None
    file_path = 'bad-words/%s.txt' % lang
    if os.path.isfile(file_path):
        raw_pattern = r'\b('
        with open(file_path) as profanity_file:
            for line in profanity_file:
                raw_pattern += line.rstrip() + "|"
        raw_pattern = raw_pattern[:-1] + r')\b'
        result = re.compile(raw_pattern)
    return result

def containsProfanity(pattern, sentence):
    if pattern is None:
        return False
    else:
        m = re.search(pattern, sentence)
        if m:
            #sys.stderr.write('Found profanity in sentence: %s\n' % sentence)
            return True
        else:
            return False

max_tokens = 100
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])





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()
        id_orig = ids.readline().strip()
        src_line_lem = src_lem.readline().strip()
        trg_line_lem = trg_lem.readline().strip()
        src_token_count = len(src_line_lem.split())
        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))):
                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))
    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))