Consider using `tok` as tokenizer; faster and more customizable

[kootenpv]

Simple example would be to import word_tokenize from tok instead of from nltk.

See: https://github.com/kootenpv/tok

[soaxelbrooke]

Hey @kootenpv, thanks for the suggestion! Care to put up a PR?