Syntactic predictability influences duration

Abstract

Building on work by Gahl and Garnsey, 2004, this paper demonstrates that speakers "buy time" during the planning of upcoming low-probability syntactic structures by producing prosodic boundaries with longer duration before low-probability than before high-probability structures. Subject extraction cleft sentences ("It was Edward who (t) loved Lucy.") are more common in corpora than object extraction cleft sentences ("It was Edward who Lucy loved (t).") (Roland et al., 2007), and are also easier to process (e.g. Gibson, 1998). The duration of the clefted constituent ("Edward") was measured in planned productions of subject- and object-extraction clefts in English. In order to disentangle the probability of each structure from its difficulty level, the probabilities were manipulated within the experiment through training. Participants read aloud: First, two SE and two OE clefts; second, eight SE or eight OE clefts; and finally, another two SE and two OE clefts. Before training, the clefted constituent was longer in OE clefts (mean 407ms) than in SE clefts (370ms, t=2.4, p=.02). This difference was no longer present after OE training (OE: 385ms, SE: 397ms), but was still present after SE training (OE: 448ms, SE: 388ms). © 2013 Acoustical Society of America.