Rouge evaluator produced `0` score for the same prediction and reference in Chinese.

I am using rouge evaluator for assessing my translation outputs, but the score is not correct as I know.
An example is like this:

both the `prediction` and `reference` are `length=9` strings. They are totally the same, but the `rouge-score` is all **zero``, while the correct one should be `{'rouge1': 1.0, 'rouge2': 0.0, 'rougeL': 1.0, 'rougeLsum': 1.0}`.