End-Weight from the Speaker's Perspective

Explanations of the tendency to put long, complex constituents at the ends of sentences (end-weight) usually take the listener's perspective, claiming it facilitates parsing. I argue for a speaker-oriented explanation of end-weight, based on how it facilitates utterance planning. Parsing is facilitated when as much tree structure as possible can be determined early in the string, but production is easiest when options for how to continue are kept open. That is, listeners should prefer early commitment and speakers should prefer late commitment. Corpus data show that different verbs exhibit different rates of word-order variation that are systematically related to differences in subcategorization possibilities in just the way predicted by a strategy of late commitment. Thus, a speaker-based account of lexical preferences in word ordering does a better job of explaining variation in weight effects than a listener-based account.