Continual Learning for Recurrent Neural Networks: an Empirical Evaluation