| import pickle | |
| import string | |
| import fire | |
| import nltk | |
| def main(target_lang='english', data_path='data/data.pkl'): | |
| key_hand = 'txt_en' if target_lang == 'english' else 'txt_ru' | |
| key_auto = 'txt_ru2en' if target_lang == 'english' else 'txt_en2ru' | |
| tokenizer_en = nltk.tokenize.TweetTokenizer(preserve_case=False) | |
| data = pickle.load(open(data_path, 'rb')) | |
| references = [] | |
| hypotheses = [] | |
| for doc in data: | |
| references += [ | |
| [[tok for tok in tokenizer_en.tokenize(vars(edu).get(key_hand)) if not tok in string.punctuation]] | |
| for edu in doc.edus] | |
| hypotheses += [[tok for tok in tokenizer_en.tokenize(vars(edu).get(key_auto)) if not tok in string.punctuation] | |
| for edu in doc.edus] | |
| print('BLEU score =', nltk.translate.bleu_score.corpus_bleu(references, hypotheses)) | |
| if __name__ == '__main__': | |
| fire.Fire(main) | |