They don’t process words as unified tokens for something like an LLM, but they do process them as multi-letter encoding, like byte-pair encoding or more advanced techniques.