I don't have the references right at hand, but there's a current theory of
linguistics that humans' language processors are innately capable of making
certain distinctions like consonants vs vowels, and recognizing
certain semantic categories like actor vs object. (And many, many more in
each category.) These classifications are ordered by importance to produce
rules of morphology, grammar, etc. But natural languages vary wildly in
the importance ranking; e.g. in Chinese and Lojban words must end in vowels
(or .) and you can't split nonconforming text into words, while in English
the vowel-consonant distinction at word end is at the bottom of the
importance list.