Source Code for Module featuregenerator.bleu.bleugenerator

 1  ''' 
 2  Created on 07.10.2011 
 3   
 4  @author: Eleftherios Avramidis 
 5  ''' 
 6   
 7  from featuregenerator.featuregenerator import FeatureGenerator 
 8  from nltk.tokenize.punkt import PunktWordTokenizer  
 9  from tempfile import mktemp 
10   
11  from os import unlink  
12  import os 
13  import subprocess 
14  import sys 
15  import codecs 
16  import bleu 
17   
18   
19     
20   
21 -class BleuGenerator(FeatureGenerator): 
22      ''' 
23      Provides BLEU score against the reference 
24      ''' 
25       
26 -    def get_features_tgt(self, target, parallelsentence): 
27          """ 
28          Calculates BLEU score for the given target sentence, against the reference sentence 
29          @param simplesentence: The target sentence to be scored 
30          @type simplesentence: sentence.sentence.SimpleSentence 
31          @rtype: dict 
32          @return: dictionary containing Levenshtein distance as an attribute  
33          """ 
34          target_untokenized = target.get_string() 
35          try: 
36              ref_untokenized = parallelsentence.get_reference().get_string() 
37   
38              bleu_value = bleu.score_sentence(target_untokenized, [ref_untokenized]) 
39              return {'ref-bleu': '{:.4}'.format(bleu_value)} 
40          except: 
41              return {} 
42   
43   
44   
45 -class CrossBleuGenerator(FeatureGenerator): 
46      ''' 
47      Provides cross-BLEU score of the current target sentence against the others 
48      ''' 
49       
50 -    def get_features_tgt(self, translation, parallelsentence): 
51          current_system_name = translation.get_attribute("system") 
52          alltranslations = dict([(t.get_attribute("system"), t.get_string()) for t in parallelsentence.get_translations()]) 
53          del(alltranslations[current_system_name]) 
54          references = alltranslations.values() 
55          bleu_value = bleu.score_sentence(translation.get_string(), references) 
56          return {'cross-bleu': '{:.4}'.format(bleu_value)} 
57           
58               
59               
60           
61           
62           
63           
64       
65   
66   
67  #    def bleu(self, translation, reference): 
68  #         
69  #        translation = " ".join(PunktWordTokenizer().tokenize(translation)) 
70  #        tfilename = mktemp(dir=u'/tmp/', suffix=u'.tgt.txt') 
71  #        tfile = codecs.open(tfilename, 'w', 'utf-8') 
72  #        tfile.write(translation) 
73  #        tfile.close() 
74  #         
75  #        reference = " ".join(PunktWordTokenizer().tokenize(reference)) 
76  #        rfilename = mktemp(dir=u'/tmp/', suffix=u'.ref.txt') 
77  #        rfile = codecs.open(rfilename, 'w', 'utf-8') 
78  #        rfile.write(reference) 
79  #        rfile.close() 
80  #         
81  #        ofilename = mktemp(dir=u'/tmp/', suffix=u'.out.txt') 
82  #        ofile = codecs.open(ofilename, 'w', 'utf-8') 
83  #         
84  #        path = os.path.dirname(__file__) 
85  #        bleupath = os.path.join(path, "bleu") 
86  #        print bleupath 
87  #        subprocess.call([bleupath, "-s" , "-p", "-S", "-r", rfilename, tfilename], stdout = ofile) 
88  #        ofile.close() 
89  #        ofile = codecs.open(ofilename, 'r', 'utf-8') 
90  #        output = ofile.readline() 
91  #        output = float(output) 
92  #        return output 
93  # 
94  # 
95