Choice between different but equally informative proposals of 'deep' (or 'underlying') structures appears to be arbitrary as well as undesirable. The appearance of rivalry among them is due to misrepresenting, in terms of alternative rules (constraints), what are in fact compatible tendencies. Realization of these tendencies, unlike the operation of rules, depends on specific lexical choice and context. Semantic tendencies operate within the framework of grammatical rules, but are no part of it. It is the freedom of ad-hoc choice among them, rather than any generative mechanism, that constitutes the truly creative aspect of language. Anyone surveying present-day linguistic studies will find that linguists of the most diverse persuasions tend to view syntactic constructions in some such way as Saussure viewed the linguistic sign-Janus-like, facing in two directions: each sentence or phrase being endowed with, on the one hand, a surface structure adapted to its phonetic representation, and, on the other, a deep structure which is adapted to its semantic representation. Clearly, there is something uncontroversial about this view; in SOME sense, a sentence or phrase-like any sign-has meaning as well as form. Yet how exactly to make sense of this truism is more problematic; more especially, the assumption that we can make sense of the two aspects of a sentence or phrase, its form and its meaning, by ascribing two transformationally related structures to it, seems dubious. Through being divided and related in this way, both seem to suffer from arbitrariness: the surface structures for lack of functional semantic controls, and the deep structures for lack of decisive formal evidence. As to surface structures, we are told explicitly that they are inadequate; and everyone will agree. But we are also told that their 'labeled bracketing' (which is, on the whole, traditional IC-analysis) is more or less uncontroversial. The deficiencies are not to be amended or replaced; they are to be cured by SUPPLEMENTATION: by deriving those labeled bracketings from more informative deep structures. This uncritical acceptance of traditional IC-analyses-so very different from the attitude of those who first proposed them-is of course difficult to justify. The placing of the brackets is very largely inexplicit guesswork, and the labeling very largely dogmatic orthodoxy-not to speak of the arbitrary assumption that constituents must be continuous, for the purely notational reason that discontinuous constituents would not sit comfortably on the branches of tree diagrams. All this seems to be so unprincipled and obscure as to be incapable of a cure by mere supplementation. For the present, however, I propose to ignore this side of the dichotomy, and turn to an examination of Janus's other face. (For some further discussion of the problem of 'surface structure', cf. Haas 1972, 1973.) It seems to be generally accepted that the supplementary information provided by deep structures is primarily semantic information. The much discussed difference between those who, like Chomsky, stipulate a further interpretive 282