In a series of seven studies, this paper examines acoustic characteristics of the spontaneous speech production of the English dative alternation ("gave the book to the boy/ the boy the book") as a function of the probability of the choice between alternating constructions. Probabilistic effects on the acoustic duration were observed in the acoustic signal at the choice point (the first word that commits the speaker to one of the alternatives), before the choice point, but not after the choice point. These findings speak in favor of the simultaneous operation of production mechanisms consistent with both information-smoothing theories and availability-based models of speech production: they are incompatible with a number of competing theoretical accounts. Finally, we outline the statistical modeling procedure of multimodel inference suitable for addressing our multiple working hypotheses and the ultimate question of the explanatory role of probability. (Contains 13 tables.)