Does an LSTM forget more than a CNN? An empirical study of catastrophic forgetting in NLP