Processing Accounts for Superiority Effects Ivan A. Sag Stanford University Philip Hofmeister UC San Diego Inbal Arnon Neal Snider Stanford University Florian Jaeger University of Rochester March 9, 2008 1 Introduction At least since Kuno & Robinson 1972, it has been widely assumed within generative discussions that contrasts like (1a) vs. (1b) are to be explained by principles of gram- mar, indeed principles of Universal Grammar, such as Chomsky’s (1973) ‘Superiority’ condition or the Minimal Link Condition of Chomsky (1995: 296): (1) a. Who broke what? b. *What did who break ? However, in an important early paper, Bolinger (1978) argued that in the right context, examples like (2) are in fact grammatical, as are examples like the following: (2) I know what just about everybody was asked to do, but what did who (actually) do ? These Superiority ‘violations’ (henceforth SUVs) were subsequently explained away by Pesetsky (1987), along with examples like (3) [Karttunen (1977)], as the result of a distinct grammatical mechanism which he called ‘D(iscourse)-Linking’. (3) Which newspaper did which student read ? To this day, most generative discussions of wh-constructions 1 assume that Superiority Effects (SEs) are the norm, to be explained by principles of UG, and that SUVs are a sec- ondary phenomenon, to be analyzed via D-Linking, or some other mechanism peripheral to the central concerns of grammatical theory. In the alternative approach to SUVs considered here, examples like (1b) and (2) are all treated as fully grammatical and are analyzed uniformly by the grammar of wh-interrogatives, which includes neither a SE-inducing condition nor any D-Linking apparatus. The reduced acceptability of examples like (1b), examined more closely below, is explained in terms of the interaction of factors known on independent grounds to contribute to processing difficulty, which in turn leads to degraded acceptability. 2 1