LSTM shakespeare model generates two spaces instead of letter 'u'

[cwalther]

Using ml5.LSTMGenerator with the shakespeare model generates text in which everywhere a lowercase letter u should appear, two space characters appear instead. Uppercase letters U are missing completely.

This is also seen in the vocab.json file, which contains no entries for "u" and "U", but instead ones with a double space and an empty string, respectively. I tried changing them there, but that did not change anything in the outcome, apparently the data does not come from there but from somewhere in the binary files (whose format I am not familiar with).