Frame semantic control of the coordinate structure constraint

Frame Semantic Control of the Coordinate Structure Constraint George Lakoff University of California, Berkeley When Ross first discovered the coordinate structure constraint twenty years ago, he almost immediately discovered a case where it fails. It is a well-known case in which extraction is possible from the second conjunct in (1). (1) What did Harry go to the store and buy? Goldsmith (1985) cites the converse case, in which extraction is possible in the first conjunct of (2) How much can you drink and still stay sober? In an attempt to save the coordinate structure constraint, [observed that the and in (1) is not a simple conjunction and that the semantic relation between the two clauses is the relation that holds between a main clause and a purpose clause (see Ross, 1986, p. 103). I proposed that (1) was not a true conjunction syntactically, but that, (in the spirit of the generative semantics of the day) its syntax followed its semantics, and that it functioned essentially like the sentence What did John go to the store to buy? Goldsmith, also attempting to save the coordinate structure constraint, argues similarly that the semantic relationship between the two clauses is like that between a main clause and an adversative clause as in How much can you drink while still staying sober? Goldsmith suggests that the semantic relationship between the clauses forces a qreanalysisq so that the syntactic properties of the sentence accord with its semantic properties. Though different in detail, Goldsmith’s analysis was very much in the same spirit as mine. I now believe that Goldsmith and I were both right in our insights concerning the correlation of syntax and semantics, but wrong in our impulse to save the coordinate structure constraint. I will argue that, in (1) and (2), and is a true con- junction and that there is a true coordinate structure. Extractability, rather than being a purely syntactic matter, depends upon the framing (in Fillmore’s sense) of the sentence in context. As a consequence, it will follow that the coordinate struc- ture constraint, as a pure syntactic constraint, does not exist. Thus, syntactic solutions, say, of the sort provided in GPSG using slash categories cannot work (see Gazdar et al, 1985). Moreover, any theory of syntax that requires that the coordinate structure constraint exist in the syntax is simply incorrect. The argument has two parts: (I) I will show that across-the-board extraction of the sort requiring true conjuncts occurs with the so-called in order to and despite senses of and, and that there is no purely syntactic coordinate structure constraint. (II) I will argue that the data can only be accounted for by a frame semantics (in Fillmore’s sense). In 1966, when Ross and I worked together on the constraints that he reported on in his dissertation, we came upon sentences like (1) above. In our haste to explain them away, we failed to apply the most basic test that any first-year syn- tax student learns to apply -- iteration. We simply never checked to see whether multiple across-the-board extractions were possible for such cases. As it turns out,