diff --git a/python/paddle/dataset/wmt16.py b/python/paddle/dataset/wmt16.py
index 39c6e78b9b431d0747118977eca3e9b476488344..4a0c1f8cb663ec105030ac2c5a70c5f906cf6d12 100644
--- a/python/paddle/dataset/wmt16.py
+++ b/python/paddle/dataset/wmt16.py
@@ -78,7 +78,7 @@ def __build_dict(tar_file, dict_size, save_path, lang):
                     six.iteritems(word_dict), key=lambda x: x[1],
                     reverse=True)):
             if idx + 3 == dict_size: break
-            fout.write("%s\n" % (cpt.to_text(word[0])))
+            fout.write("%s\n" % (cpt.to_bytes(word[0])))
 
 
 def __load_dict(tar_file, dict_size, lang, reverse=False):