Can I classify laugh words with char level tokenization?

Question

I want to use Keras\' tokenizer\'s char_level capability to train a model to classify if a word is a laugh or just a regular word, like "hello" (regul