Interface extensions: Freezing as a probabilistic phenomenon

This paper shows that freezing effects are graded rather than categorical, and that different kinds of freezing are not equally strong. Building on Hofmeister et al. (2015: 470), I argue that freezing effects are at least in part caused by their extremely unusual structure, with two disparate foci governed by the same verb. By being inconsistent with comprehenders' expectations about the distribution of gaps, such constructions likely create a processing conflict between what is expected and the actual input. Experiment 1 suggests that such expectations are malleable, given that the oddness of extracting from an extraposed phrase disappears by virtue of making such constructions as likely as their non-extraposed counterparts. Experiments 2 and 3 suggest that the oddness created by crossing extraposition and extraction paths also disappears, but at a much lower rate. I propose that the latter constructions are more improbable and therefore worse than the former because (a) they are preempted by simpler and more likely alternative (local) parses (Fodor 1978) in which the point of retrieval and integration does not coincide with the point of reanalysis (Hofmeister et al. 2015), (b) involve crossing non-local dependencies (which are independently known to be more difficult than non-crossing dependencies (Fodor 1978), and therefore bound to be rarer), and (c) have disparate foci and therefore atypical pragmatic requirements (Huck and Na 1990; Bolinger 1992).