not specifically a pytorch question, but does the vocab size matter when training RNNs? I started with a character level rnn, but Im wondering if I move to longer words/specialized vocabulary, will it improve the model performance? From what I understand of RNNs, using longer length vocab words will mean that the RNN layers can learn longer sequences because there is more information embedded in the RNN. Are there any papers that go over this kind of topic that I should read?