Non-projectivity is an important theoretical and computational concept that has been investigated extensively in the dependency grammar/parsing paradigms. However, from a human sentence processing perspective, non-projectivity has received very little attention. In this paper, we look at existing work and propose new factors related to processing non-projective configuration. We argue that (a) counter to the claims in the psycholinguistic literature (Levy et al, 2012), different aspects of prediction maintenance can lead to higher processing cost for a non-projective dependency, (b) parsing strategies can interact with the expectation for a non-projective dependency, and (c) memory (re)activation can explain processing cost in certain non-projective configurations.
展开▼
