diff --git a/hunvec/datasets/tools.py b/hunvec/datasets/tools.py
index 593d15b..a1a19ea 100644
--- a/hunvec/datasets/tools.py
+++ b/hunvec/datasets/tools.py
@@ -11,7 +11,7 @@ def replace_numerals(w):
         i += 1
         begin = w[:len(matched.groups()[0])]
         end = w[len(matched.groups()[0]) + len(matched.groups()[1]):]
-        w = '{}__num__{}'.format(begin, end)
+        w = u'{}__num__{}'.format(begin, end)
         matched = num_pattern.match(w)
     return w