Does the tokenizer for tweetnlp cast everything to lowercase? On the demo, I see similar sentiment for uppercase and lowercase phrases. Thanks!