The ‘balanced’ or ‘generative semantic’ view of language is assumed to be correct, but is seen as giving rise to certain other fundamental questions. One is the question of directionality. Contrary to what has been the dominant view, it is argued here that language exhibits a directionality from semantics to phonetics, and several kinds of evidence are presented. The discussion focuses on the directionality of well-formedness, the directionality of the speech act, and the directionality of ambiguity. The last topic leads into a discussion of paraphrase. True paraphrases, in the sense of two or more surface structures which reflect a single semantic structure, are seen to be relatively uncommon. The belief that paraphrase is a widespread phenomenon is attributed to an insensitivity to meaning differences, to an unfortunate reliance on the truth-value criterion, and to the fact that different semantic structures may converge on meanings which are—at least in some contexts—equivalent. The discussion of this last point relies on a distinction between semantic structure and meaning which is seen as analogous to the distinction between phonetic structure and sound. It is left as an unanswered but important question whether semantic structures should be formally represented in such a way that their equivalence of meaning, even if contextually limited, is deducible.