Continual Learning of Recurrent Neural Networks by Locally Aligning Distributed Representations