The smallest unit of **meaning** in words. E.g. In "unhappiness" there are 3 morphemes, "un", "happy", "ness" Some models try to use tokenization based on morphemes. Particularly good for low resource languages