Does a theory of language need a grammar? Evidence from Hebrew root structure