Guest speaker: Dave Kush (Norwegian University of Science and Technology)

We are delighted to welcome Dave Kush, who is a faculty member in linguistics at the Norwegian University of Science and Technology, based in Trondheim. He earned his Ph.D. from the University of Maryland in 2013, and his research focuses on syntax, psycholinguistics, and typology. His talk, "How does the grammar guide incremental prediction during sentence processing?", will be held via Zoom meeting on Tuesday, May 12, from 3:15 to 4:45 PM. Note that you should register to attend via the link in the email.

Previous work (e.g. Stowe 1986) has argued that comprehenders pursue an active strategy when
processing incomplete wh-filler-gap dependencies such as (1):

(1) Frida wondered who...

 Comprehenders appear to predictively assign a gap/trace position for the dislocated wh-filler "who" in (1) before they receive evidence for the true gap position. In this talk I investigate (i) whether such instances of pre-emptive structural commitment reflect a more general strategy of active prediction across dependency types and (ii) how to best characterize the underlying motivation for active prediction. To this end I discuss recent studies that investigate the role of active prediction in the processing of cataphoric, relative-clause, and embedded question dependencies in English and Norwegian. In the first half of the talk I show that comprehenders incrementally predict yet-to-be-seen referents during the online resolution of cataphoric pronouns. I suggest that these results motivate a general, cross-dependency characterization of the mechanisms underlying active prediction. In the second part of the talk I consider how animacy information impacts predictions for gaps in filler-gap dependencies. Results suggest that comprehenders predict gaps for animate fillers, but not for inanimate fillers. I argue that these results are incompatible with most models of expectation-driven dependency parsing, including diagnostic utility models (Wagers and Phillips 2014) and models that treat expectation as arising via (rational) inference over statistical regularities in comprehenders’ language experience (e.g. Levy 2008).