In the context of Natural Language Processing $($NLP$),$ why is it stated that tokenization is "LOOSELY based on words"?

Tokenization is primarily focused on sentence segmentation.

Tokenization may split text into units that are not always equivalent to words due to factors like contractions, hyphenation, or punctuation.

Tokenization is a term exclusively used in computer programming and not related to NLP$.$

Tokenization is an unrelated process and doesn't involve language units.