Bytes: 2 bytes counts a Korean or Japanese character the way EUC-KR and Shift_JIS store it — and the way most forms and SMS limits still count it. 3 bytes is what UTF-8 actually stores: “こんにちは” is 5 characters, 10 bytes by the first and 15 by the second.

Word bounds: words are split on spaces and punctuation, so “Hello, world!” counts as 2.