What type of tokenizer is used?